amd Applications

A-JEPA AUTOMATA

A formally-verified AI agent platform where Bright Data feeds live web intelligence — competitor signals, filings, pricing — and proof audits every decision before it lands in an append-only ledger. Built for enterprise trust at scale.

Apohara CONSILIUM — Agent Governance OS

The Agent Governance OS for autonomous AI in regulated industries. 9-vendor adversarial ensemble + Z3 formal proof + RFC 3161 court-grade audit chain. DORA + EU AI Act Art. 14 ready. Open-source Apache-2.0. Live demo at apohara.dev/consilium.

Autonomous AI Agents for CRISPR Diagnostics

An intelligent autonomous AI agent system built to automate CRISPR diagnostics and optimize clinical workflows, ensuring precision healthcare reaches underserved communities in Africa.

MusKent Commerce OS

MusKent is an autonomous commerce platform using AI agents to analyze business signals, reason over context, and execute workflows across revenue, sales, automation, and marketplaces with structured intelligence and resilient fallback systems.

Apohara PROBANT — Cross-AI Code Verifier

Open-source cross-AI verification: Gemini writes code, a 12-vendor adversarial ensemble audits it. Z3-proven INV-15 KV-cache isolation. Every verdict HMAC-signed.

klarixa-tricortex-amd-hackathon

An asynchronous AI orchestration core built with pydantic-ai and the Model Context Protocol (MCP). Features deterministic tool-calling, secure data boundaries, and in-memory latency emulation. Cross-validated on AMD clúster hardware configurations.

Fuzzy RAG

Fuzzy RAG is an explainable Retrieval-Augmented Generation system that combines vector similarity with fuzzy logic to provide context-aware document ranking.

Nexus

Nexus uses IBM Bob's full-repo context to extract Decision Provenance Records, builds a semantic Causal Temporal Graph in Neo4j, monitors assumption decay via blast-radius scoring, and surfaces org knowledge risk before it becomes an incident.

CancerLens AI — Cancer Detection AI

AI-powered cancer detection platform built entirely with IBM Bob. Upload any medical scan and get a complete clinical report with cancer staging, survival statistics, and specialist recommendations in seconds — not weeks.

Atlas Sanctum: Autonomous Enterprise Command AI

Atlas Sanctum is a multi-agent AI command system that transforms enterprise data into autonomous decisions and real-world actions using intelligent reasoning, multimodal analysis, and collaborative AI workflows.

Vision-Link AI: Genomic & CRISPR Agent

An autonomous AI-Agent system leveraging IBM watsonx.ai for high-precision genomics and CRISPR analysis to address health challenges in Sub-Saharan Africa.

Q-Optima - Autonomous Quantum Logistics Agent

Solve NP-Hard logistics in 30 seconds using Quantum AI, Speechmatics STT, and live weather/traffic/carbon enrichment at zero cost.

AluminatiAI

AluminatiAI GreenTune is an autonomous AI energy optimization platform that uses multi-agent systems and real-time GPU telemetry to reduce the cost, power usage, and carbon footprint of enterprise LLM training workloads.

Private ML system

Privatising ML systems by applying mathematical techniques like differential privacy. This project implements Differential Privacy (DP) atop the Veea Edge Platform to create a decentralized, secure-by-design machine learning architecture

Project Doomsday

A multi-agent AI engine that stress-tests stock valuations against Black Swan events using adversarial Bear/Bull/Judge debates, geographic risk mapping, contagion cascade modelling and transparent 5-path distressed valuation — powered by Gemini 2.0 Flash

ACMI: The Universal Memory Layer for AI Agents

Three keys per entity. One memory protocol. Every framework. Every model. One audit trail via Veea Lobster Trap. Production-proven across 9 agents over 30+ days.

TrafficOS: Gemini-Powered Mobility Intelligence

TrafficOS uses live traffic data, predictive routing, SARIMAX forecasting, and Gemini-powered ARIA reports to help cities and fleets optimize fuel, time, congestion risk, and delivery reliability.

Adaptive Cognitive Runtime:Autonomous Engineering

CogniRoute is an AI orchestration that allocates reasoning acros agents instead of relying on one monolithic model. Heavyweight reasoning models handle planning & verification, while lightweight workers execute scoped implementation tasks.

one-shot music video creator

Enter a topic, and a style, and then lyrics and a video is generated automatically

Theorem

A hand-tuned set of GPU kernels for the AMD MI300X. Today: re-wrote the gated DeltaNet WY-transform from an O(C²) elementwise loop into two MFMA GEMMs — 2.4× speedup. Wavefront-aware num_warps + L2 reordering + persistent kernels make CDNA3 sing.

AssureSearch

AI-powered legal document intelligence platform that analyzes contracts, extracts clauses, identifies risks, generates summaries, and enables semantic legal search using AMD GPU-accelerated agentic workflows and open-source LLM infrastructure.

Autonomous Flight Commander

Real-time autonomous drone swarm simulator with PID physics in C++, LLM strategic orchestration via Gemini AI, TCP telemetry bridge, and live web dashboard — powered by AMD MI300X and ROCm.

GuardianRail

Representation-aware action firewall for open-weight agents: watches Gemma Scope features, blocks risky operations, and logs why each action was allowed or escalated.

ROCmify-AI

AI-powered copilot for migrating CUDA/PyTorch workloads to AMD ROCm.

Parasite AGI first AGI

Its an agi it grows always it can make tools and use the it can fit any where It learns like a child It simulates the brain and beoynd It compete whole agents It can do anything online and use open sourses Its the begining of future

MOVIMENTO

Movimento turns natural language into controllable 3D human and robot motion by combining Qwen agentic planning with LLM2Vec semantic encoding and Kimodo diffusion generation in a production-minded AMD-ready workflow.

Uplan: A Stateful Deep-Agent Document Intelligence

We present Uplan, a privacy-first, deployed deep-agent platform that treats every immigration application as an adversarial interrogation target.

Zero-Noise

A powerful Security Operations Center (SOC) dashboard powered by AMD-optimized modular agentic architecture. ZERO-NOISE intelligently analyzes massive log volumes to identify and present only critical anomalies and security threats.

Rocm Potter

Complete ROCm harness featuring the ROCm compiler, profiler, specialized skills, and intelligent agents. Converts CUDA code to HIP and optimizes it for performance on MI-300x platforms.

Apohara-ContextForge

Shared context compiler for multi-agent LLM systems. Cuts token usage 79%, VRAM 68%, TTFT 7.8x. Implements TokenDance, JCR Safety Gate & KVCOMM on AMD MI300X. 310 tests passing, 15/15 benchmarks.

Factory Flow-an autonomous predictive parts agent

FactoryFlow is an autonomous predictive maintenance and parts procurement agent for small-to-medium manufacturers

Med Trace AI

MedTrace AI turns fragmented patient records into a temporal knowledge graph so clinicians in low-resource settings get memory, grounded evidence, and research-style prompts—without replacing clinical judgment or making diagnoses.

Ken: The Real-Time Co-Listener

Ken hears every word. You understand every one. Ken is a free, self-hostable, open-source AI co-listener that gives real-time, explainable, personalized support during immigration, legal, and medical conversations — built on AMD ROCm.

Novel Storyteller

A stateful neuro-symbolic AI narrative engine for stories that uses structured state as the source of truth. It generates stories through DAG-based scene execution, selective context propagation, incremental verification, and validated state updates.

Verixa — Runtime Compliance Control Plane

Verixa is a runtime control plane between AI agents and regulated systems. It applies a tool-call firewall, signed Rego policies, and an Ed25519-signed audit chain. Annex IV-aligned. Built on AMD MI300X via vLLM-on-ROCm.

Strike Lab

Strike Lab is an AI-powered biomechanical platform that transforms smartphone footage of football strikes into professional-grade performance reports.

TurtleTalk

TurtleTalk is a voice companion that helps kids reframe hard moments into actionable missions and helps parents build a deeper connection with their kids. Our mission is to make education easy and accessible for everyone.

Cardiocore

CardioCore analyses ECG images and echocardiogram videos using specialist AI models on dual AMD MI300X GPUs, computes a HEART risk score, and generates a validated FHIR R4 diagnostic report in under 5 seconds.

AMD-Link: Autonomous PCB Routing for Ryzen

AMD-Link is a four-stage multi-agent pipeline (ingest → extract → route → check) that turns a natural-language prompt and a raw .kicad_pcb into a routed board, powered by a local-Ollama RAG layer with grounded per-field extraction and self-RAG critique.

Nullstate-CLI

Autonomous purple-teaming CLI for infrastructure-as-code sandboxes.

Ukhozi - Multi Agent Disaster Response Simulation

Worker-authoritative simulation for multi-agent aerial and ground vehicle experiments.

Preserving Heritage with AI-Driven PDF Restoration

Project aims to develop a platform that converts old, yellow, stamped, or poor-quality PDF documents into clean, professional, white-background PDF files. It will remove blank pages, enhance quality, remove stamps and provide easy-to-use interface.

AIPP - Autonomous AI Pair Programmer

A voice-first AI assistant that uses vision to monitor the screen and autonomous code injection to help developers code hands-free using AMD hardware acceleration.

Verilog Fine-Tune: Chip Design AI on AMD MI300X

Fine-tuned Qwen 2.5 Coder 7B on a proprietary Verilog/VHDL/SystemVerilog corpus for chip design code generation. Built at the AMD Developer Hackathon SF (May 9-10, 2026).

Credora - AI Credit Scoring for Informal Workers

Credora is an AI-powered credit scoring platform that turns everyday financial evidence – bank SMS, receipts, payment screenshots – into a portable, transparent credit profile for informal workers who are invisible to traditional banks.

Cloud Pharmacy - Inventory and LLM Assistance

A cloud-ready prototype for managing pharmacy stock, validating medicine images, training a pill detector, and connecting a pharmacist-focused AI

BrainOS - The knowledge layer for AI agents.

BrainOS turns Slack, docs, PDFs, and diagrams into a reconciled knowledge graph. Live Slack integration answers @BrainOS queries in-channel and exports portable SKILLS.md context for Claude, GPT, Cursor, and other AI agents.multi-modal on AMD MI300X.

PitchSideAI

Real-time football intelligence for fans and broadcast teams, powered by agentic match research, tactical vision Q&A, and AMD-ready ROCm/vLLM infrastructure.

SafeGuard AI

SafeGuard AI is a multi-agent system for the AMD Hackathon. Using CrewAI, it analyzes IoT sensor data to detect anomalies. A safety agent then drafts repair protocols. Optimized for AMD Instinct MI300X GPUs and the ROCm stack.

Enterprise Architecture Strategy Optimizer

Transforms a 10min EA questionnaire into governance-grounded, Jira-ready roadmaps. Agentic pipeline: Graph RAG (44 domains,1,400+ capabilities) → DRL prioritisation → Qwen-72B on AMD MI300X via ROCm → compliance verification. Built on LangGraph + Neo4j.

MindForge AI: Mental Health Intelligence

Continuous mental-health intelligence from device, adherence, caregiver, and patient check-in signals.

SafeSight

AI safety inspector for worksites. Upload construction footage, auto-detect PPE violations with YOLO, browse timestamped evidence frames, and ask natural-language questions about inspection results. Powered by Qwen on AMD GPU cloud.

AMD Smart Product Assistant API

AI-powered Smart Product Assistant that delivers real-time product recommendations, conversational support, and personalized insights using LLM technology and scalable APIs.

SentinelAI — Autonomous Multi-Agent AI SOC

SentinelAI is a hackathon-grade, production-shaped autonomous Security Operations Center. It continuously ingests telemetry through collector agents, normalizes and enriches events, runs multi-modal detection.

Multi-Agent Clinical Reasoning Framework

A multi-agent AI framework for clinical reasoning that evaluates iterative decision-making workflows using coordinated AI agents to solve, critique, and refine diagnostic reasoning on benchmark healthcare question-answering tasks.

Pathology Report Translator

Fine‑tuned Llama 3.1 to convert complex pathology reports into clear, empathetic patient explanations (8th grade reading level). Built with AMD MI300X GPU.

BiasharaMind AI-Powered Business Consultant

BiasharaMind is an AI-powered business consultant for SMEs, turning business assessments into clear diagnostics, priorities, growth roadmaps, and practical reports to help owners make better decisions.

Vow AI

A real-time shared commitment and memory layer for meetings

AI-network-incident-response-agent

AI-powered multi-agent network incident response platform using LLMs, RAG, and AMD GPUs to analyze logs, detect security/network issues, perform root cause analysis, and provide automated remediation insights in real time.

ATLAS: Autonomous Travel Intelligence System

ATLAS is an AI-powered autonomous travel intelligence platform that uses agentic workflows to generate personalized itineraries, smart recommendations, and real-time travel insights for seamless trip planning.

MUDRAIS: Semantic Matchmaking Engine

Multi-tenant SaaS that converts Discord user profiles into 2048-dim semantic vectors via AMD-accelerated OSS 120B and Qdrant, enabling privacy-isolated partner matching across roleplay, book clubs, and gaming communities no custom infrastructure required

Memora Vision

Memora Vision transforms camera footage into searchable, conversational memory with AI-powered event recall, object tracking, and real-time intelligent alerts.

SynapseOS — AI Agent Civilization

5 expert AI agents (PM, Developer, Critic, Finance, Security) debate any business idea in real-time using Qwen2.5 on AMD MI300X GPU via vLLM, delivering structured analysis with GO/NO-GO decision and voice summary.

Psycholinguistic CV Analyst

An AMD-optimized, fine-tuned Qwen agent using PEFT/LoRA to perform psycholinguistic CV analysis and simulate MBTI/OCEAN team dynamics. It leverages ROCm and vLLM on MI300X to predict corporate synergy and cultural alignment at high speed.

IaC SecFix Multi Agent

IaC-SecFix is a scanner-verified AI agent for Terraform and Kubernetes that uses Qwen-based remediation, Checkov validation, safety gates, and human approval to generate reviewable security patches.

ROCmOps AI War Room

ROCmOps AI War Room is an autonomous AI-powered incident response platform accelerated by AMD MI300X GPUs and ROCm, enabling real-time infrastructure monitoring, remediation, GPU optimization, and AI-driven operational intelligence.

VitaSync

VitaSync is a private AI health platform that helps patients securely store, analyze, and query medical records while giving clinicians consent-based access to grounded insights, risk signals, and consultations through a modern patient-doctor workflow.

Shortcutsistem

A platform where AI agents learn skills from certified AI teachers. Enroll, study structured courses, chat with an LLM teacher, and earn certificates. Works with Claude Code, OpenClaw, or any REST client.

Lysos - Antimicrobial Drug Agent

Lysos is an end-to-end agentic antibiotic discovery platform. We took Google's Gemma 4 31B-it and fine-tuned it three times on AMD MI300X:

CogniCore

CogniCore transforms meetings into production-ready workflows using AI. It captures discussions, extracts decisions and tasks, then orchestrates development, governance, and execution automatically in one unified platform.

Startup Advisor AI

A specialized, fine-tuned large language model (LLM) that transforms generic AI responses into highly structured, actionable, and framework-grounded advice for startup founders

Streaming-MoE Stack for AMD

Streaming MoE on AMD: run mixture-of-experts models bigger than your GPU's VRAM. A kernel PR'd to ROCm/AITER, an MIT-licensed llama.cpp fork, and a stock OpenAI-compatible agent — running unchanged from a $700 RX 9070 XT to a $15K MI300X.

CropWhisper

CropWhisper is a multimodal AI agent that diagnoses crop diseases and nutrient deficiencies from photos, then delivers a plain-language, prioritized action plan — built for farmers worldwide who have a smartphone but no agronomist.

Autonomous Research Agent for Multi-Hop Reasoning

AI agent built on Gemma-4-E4B, trained via GRPO Reinforcement Learning for advanced multi-hop reasoning. It dynamically searches the web, executes code, and synthesizes complex data into comprehensive reports while preserving vision capabilities.

Project ARK: Autonomous Reconnaissance Kinematics

A sovereign AI disaster response protocol powered by AMD MI300X. Project ARK integrates NASA Prithvi-100M and agentic workflows to provide real-time, bilingual (English/Tagalog) tactical intelligence for Philippine emergency agencies (NDRRMC).

OtwoX African RLHF Foundry

A human-in-the-loop annotation platform for African languages on AMD MI300X. Native speakers of Sudanese Arabic, Fur, Zaghawa & Dinka correct AI outputs via text or voice to build sovereign African language datasets for RLHF training.

Vapor Deck

An AI-powered slide deck generator that streams fully-designed, web-native HTML presentations in real time — complete with animations, syntax highlighting, and vision-model layout auditing.

Live Sight

A vision assistant that learns its user. Live Sight runs Qwen3-VL-8B on AMD MI300X with on-device LoRA fine-tuning, voice corrections in Roman Urdu and English, and episodic recall over what you've seen. Built for blind and low-vision users.

VLSI Verification Co-Pilot

VLSI Verification Co-Pilot is an AI agent powered by Qwen 3.5 fine-tuned on AMD MI300X GPUs to instantly optimize Verilog hardware synthesis, eliminate latches, and identify testbench race conditions.

GreenChain — AI Supply Chain Carbon Auditor

GreenChain is a multi-agent AI system that audits supply chain carbon footprints in seconds. Upload shipping data and three AI agents extract shipments, calculate CO₂ emissions, identify hotspots, and generate a green sustainability roadmap.

Multimodal workflow memory for AI agents.

A multimodal context recorder that turns human screen workflows into searchable execution memory for AI agents.

Nzeru Core Health

An AI system that analyze geospatial data and community surveys to predict childhood/community diseases outbreak,identify root causes and generate targeted ministry of health response strategies specifically for vulnerable regions.

Neo-Edu: Agentic Administrative Intelligence

Bridging the gap between school data and autonomous action.

SupplyChain-Guardian AI

Supplychain guardian is an AI powered multi-agentic platform designed to help organizations and users proactively detect supply chain risk, coordinate mitigation workflows and support operational decision making in real time.

Enhanced the Hindi Language support in Qwen3_Omni

Enhanced Hindi support in Qwen3-Omni-30B-A3B-Instruct using AI4Bharat datasets (ShrutiLipi, Rasa, Lahaja,QnA) and a custom LLaMA Factory variant for Hindi audio understanding. Audio output training was limited due to lack of official codec support.

GhostTrace AI-Autonomous Cyberattack Investigation

GhostTrace AI is an autonomous cyberattack investigation platform powered by AMD Instinct MI300X GPUs that uses AI agents to detect threats, reconstruct attack timelines, analyze multimodal evidence, and generate forensic intelligence in real time.

Aurix — Compliance Intelligence via GraphRAG

Agentic GraphRAG system mapping startup policy docs against GDPR, EU AI Act, NIS2 & DORA using Neo4j knowledge graphs + LangGraph agents + Qwen3-4B-Thinking on AMD MI300X. Finds cross-regulation conflicts flat RAG cannot detect.

LUMEN - From topic to report

LUMEN is a multi-agent AI research assistant that takes any topic and autonomously plans, searches, extracts, analyzes, and writes a fully structured research report with citations — powered by CrewAI and AMD.

AproxResearch: AI-Powered Web3 Exploit Engine

AproxResearch is an AI security platform that analyzes smart contracts, surfaces exploitable risk patterns, and helps researchers validate findings with reproducible technical workflows.

Project Aura: Neural Forensic Intelligence Hub

Project Aura is an AI-powered forensic platform accelerated by AMD Instinct™ MI300X. It utilizes the Qwen-7B model via ROCm™ to analyze raw forensic artifacts and generate automated, structured investigative reports.

Wave Engine: Multi Agentic System

Wave Engine is an agentic app builder that generates, repairs, validates, and packages full web projects using a MAS-first autonomous repair loop, with LLM calls only when static agents cannot fix issues.

YuhoLens - Auditable Memos from Yuho Filings

YuhoLens is a 14B open-weight Japanese-finance LLM that ingests EDINET yuho filings and drafts cited equity memos through a 4-agent LangGraph composer, shipped with a reproducibility ledger and tuned for AMD ROCm inference.

IntelliGuard - Injection Defense for Enterprise AI

A 4-layer AI security pipeline detecting prompt injection attacks in real-time. Fine-tuned on 88k adversarial samples on AMD MI300X. Protects RAG systems from direct, encoded, multilingual, and poetic attacks.

Diagnosis assitant for icu team

AMD Cardio Agent reads ECG, CXR, echo, CT, vitals, and labs, then drafts a preliminary diagnosis for cardiology ICU teams — so junior doctors don't wait for the senior cardiologist on every case.

Counsel-in-a-Box: Legal AI on MI300X

Drop a deal package, get a senior-associate-grade legal memo in minutes. Five specialist AI agents analyze the entire document set in a single 250K-token context window no chunking, no RAG, no context loss. Only possible on MI300X's 192 GB HBM3.

AMD-CineMatch: Mood-Powered AI Agent

AMD-CineMatch is a decoupled AI agent that analyzes user sentiment to provide personalized Netflix recommendations. Optimized for AMD Instinct™ MI300X and ROCm™ 6.0 it demonstrates a scalable bridge between open-source models and high-performance silicon.

LifeMax-AI

LifeMax-AI is an AI execution engine that turns ANY goals into detailed campaigns, schedules them into real life, recommend suitable resources, and adapts the plan as external situations change.

Culltron Smart Port Flow

AI-assisted port operations platform that helps reduce truck queues, berth and yard bottlenecks, idling, fuel burn, and CO₂ impact through dashboards, simulation, Copilot guidance, and explainable recommendations.

CareerPilotX

CareerPilotX is an AI-powered career guidance platform that helps students discover career paths, build resumes, explore skills, prepare for interviews, and make smarter future decisions using advanced AI tools

path_to_care

Path to Care: multimodal phone triage for rural healthcare. Top-3 + R/Y/G urgency, never a diagnosis. Gemma 4 + LoRA on AMD MI300X (+7pp top-1).

Schedly

Life gets busy. Your schedule should adapt. An AI Agent scheduler that dynamically rebuilds plans around delays, unfinished tasks, fatigue, emergencies, shifting priorities, and overtime—balancing productivity, health, flexibility, and workload.

mindXtrain

MI300X-native LLM fine-tuner whose 60-second AOT probe locks attention, GEMM, and RCCL config before training — zero JIT in the loop. SFT/DPO/GRPO → Quark FP8 → OpenAI API with BLAKE3 manifests and on-chain receipts mindX agentic training on AMD

VisionGuard: Industrial AI Inspection on AMD

VisionGuard is a real-time industrial visual inspection system that uses Qwen-VL on AMD MI300X via ROCm/vLLM to detect product defects, trigger operator alerts, generate reports, and support factory-specific adaptation workflows.

SovereignAI-AMD

SovereignAI-AMD solves privacy issues by providing a private, hardware-attested AI pipeline that runs entirely on the AMD Instinct MI300X. With 192GB of HBM3 memory, we run massive models like Qwen2.5-72B in high precision locally.

DREAMWEAVE: Layered Memory Intelligence Framework

DREAMWEAVE is a cognitive memory framework a layered retrieval middleware that sits between raw data and any large language model, transforming how AI systems store, organize, retrieve, and reason over knowledge.

Agentic Reproducibility Engine

AMD-hosted multi-agent research reproducibility evaluator that reads papers, resolves arXiv/DOI/GitHub/dataset evidence, critiques gaps, and streams audit-ready reports.

RouterCore: Safe Routing for AI Agents

RouterCore fine-tunes a lightweight routing model on AMD ROCm to turn messy DevOps requests into validated workflow routes, clarifications, confirmations, or policy rejections before agent execution.

M3I: Where Does a Small LLM Fail, Layer-Wise?

Two LLMs audit a small MoE model's reasoning, identify failure-correlated layers via statistical attribution, and train LoRA-DPO targeting only those layers. Validated end-to-end on DeepSeek-V2-Lite; held-out eval pending controls.

Planner Agent

A terminal AI planning harness that turns vague software ideas into model-routed handoffs for coding agents

QNASAgent: Talk to NASA Data

QNASAgent is an AI-powered platform that transforms complex NASA satellite data into instant geospatial insights using natural language, enabling researchers, governments, NGOs, and non-technical users to access Earth observation data in seconds.

ClinSight: Agentic Multimodal Clinical AI

Multi-agent clinical AI that fuses chest X-rays, labs, and history through dual Qwen models on AMD Instinct MI300X via ROCm 7.0 and vLLM.

EchoLens

EchoLens turns photos, messages and voice notes into AI-generated interactive memory experiences that feel personal, immersive and easy to share.

CancerLens AI

CancerLens AI is a 3-agent cancer detection platform powered by Qwen2-VL on AMD MI300X. Upload any medical scan and receive a complete clinical report with cancer staging, survival statistics, and specialist recommendations in seconds — not weeks.

Frequence: Agentic Music Visualizer

An AI-powered music visualizer that uses an AMD MI300X and Qwen 2.5 to let users dynamically control visual aesthetics in real-time through natural language.

Intent machine X

An intent design engine. The intelligence layer to interface scientific research and physical infrastructure.

SchoolarBee

ScholarBee is an AI agent that helps students find reliable scholarships based on their profile. It searches trusted sources, checks key details, and ranks results by compatibility.

ROCKET — AI Performance Researcher for AMD

An AI scientist that autonomously makes any PyTorch model faster on AMD MI300X — achieved a measured 2.93× speedup on Qwen2.5-7B without human input.

RTL Repair

RTL Repair is a fine-tuned Verilog/SystemVerilog bug repair assistant that takes design intent and buggy RTL, then returns fixed RTL, a bug explanation, and a verification suggestion.

BrainSkribbl

Improve and evaluate your educational podcasts with brain-activity-optimized feedback using AI-predicted neural engagement.

Meridian: AI Engineering Platform for Teams

Meridian is an AI-first project management platform demo. It is powered by an AMD MI300X inference endpoint to provide a high-speed, context-aware AI assistant that acts as a PR Analyst, Sprint Coach, and an embedded autonomous coding agent.

AUDITIA ROCm-X: Public Contract Auditor

AUDITIA is an AI agent detecting opacity in Colombian public contracts. Built natively on JAX, Flax, and ROCm for the AMD MI300X , it uses an unsupervised Autoencoder to flag financial and semantic anomalies.

AI PC Troubleshooter

AI-powered PC diagnostic system using AMD MI300X, ROCm, and vLLM. Multi-agent CrewAI pipeline diagnoses hardware, software, and network issues with step-by-step solutions via a Next.js dashboard.

Radiology AI

An AI-assisted radiology workflow system that combines medical image segmentation with vision-language report generation to produce draft radiology reports, ready for physician review.

IMMUNISACIN AdversarialCoevolutionaryImmuneNetwork

self-evolving multilingual cyber immune system that detects threats in 40+ languages, synthesises defences through adversarial AI battle, formally verifies with Z3, broadcasts immunity across an encrypted P2P mesh-every connected org inherits protection.

FinAgent: AI Trading Research on AMD MI300X

Five specialist AI agents on a single AMD MI300X produce grounded BUY/SELL/HOLD signals with entry, stop, target, and per-agent reasoning, powered by Qwen3-14B, vLLM, and ROCm 7.2. No external LLM APIs, no subscriptions.

Nightmare Shield

A hybrid distributed NOC/SOC infrastructure running local AI inference natively on an AMD Ryzen 7, unifying edge and legacy hardware for secure, decentralized monitoring

Alonda: Real-Time Exam Cheating Detection

ALONDA is an AI proctoring system for detecting exam malpractice. It uses MediaPipe for pose estimation and a Multi-layer LSTM network to perform temporal posture analysis, accurately classifying cheating behaviors like sharing answers.

Faultline

Faultline is a multi-agentic platform to simulate how real users fail, abandon, misunderstand, and exploit software before release.

Vaara: runtime AI Act compliance kernel

Solo builder shipping AI agent governance on AMD ROCm. Vaara is an open-source Python runtime that intercepts AI agent tool calls, scores risk with split conformal prediction, and writes a W3C PROV-JSON audit trail for EU AI Act Article 14 compliance.

StudioMI300 - cinematic reel from one prompt

One prompt becomes a 30-second cinematic reel - planning, video, music, voice-over - end-to-end on a single AMD Instinct MI300X. Apache 2.0 / MIT all the way down. Qwen3.5-35B + FLUX.2 klein + Wan2.2-I2V + ACE-Step + Kokoro.

InfraAgent ReplayOps

Evidence-verified incident triage on AMD: a LangGraph agent turns noisy alerts into auditable War Room Packets with evidence IDs, owner routing, runtime truth, and human-approved recovery plans.

Dissent: AI Red Team Workspace

Dissent turns authorized AI lab, CTF, or secure product context into evidence gaps, failure chains, adversarial agent debate, and writeup-ready next steps.

JEPA News Recommendation System

A self-supervised news recommendation system using Joint-Embedding Predictive Architecture (JEPA) on the MIND dataset, benchmarked against an XGBoost baseline.

AgriSync

AI-powered crop disease diagnosis and market arbitrage for Africa's smallholder farmers — running on AMD MI300X. Snap a photo, get a treatment plan and the best market to sell at, in English and Swahili.

DynoDocs

AI architecture diagramming tool that scans a codebase, extracts verified services and dependencies, then generates an interactive component diagram with animated data flow, evidence-backed architecture insights, and static bottleneck risk detection.

DAGIntel-Multi-Agent Airflow Incident Investigator

AI-powered incident response system that compresses 4+ hours of Airflow debugging into 90 seconds using three specialized AI agents. Generates production-ready runbooks with actual code fixes, diagnostic commands, and prevention strategies.

BineuralClaw

BineuroClaw is an inference-only Audio JEPA pipeline that turns speech into 256D acoustic latents, predicts 80‑band log‑mel spectrograms, and can reconstruct waveforms with HiFi‑GAN for analysis and demos on AMD MI300X.

Forma — The Design Memory Layer for Vibe Coders

Forma is a Chrome extension that catches vague UI vocabulary in vibe coders' prompts on v0, Cursor, Lovable, and Bolt — rewriting them with canonical component terms. 7 agents on AMD MI300X: Llama 3.1 8B for real-time scoring, 70B AWQ for Memory Engine.

Orvex

Orvex is an AI inspection copilot for solar teams. It analyzes panel images and video frames, prioritizes risks, generates structured findings, and keeps every result human-reviewable with transparent evidence.

AuditEye | Sovereign Transparency

A locally hosted AI auditor powered by Qwen 2.5 and AMD MI300X. It cross-references government and enterprise procurement data against market baselines to detect price bloat and corruption securely at the edge.

SKT-OM: 13B LLM + LangGraph Agentic RAG

SKT-OM is a powerful 13B LLM integrated with SKT RAG and LangGraph on AMD ROCm. It delivers intelligent, multi-step reasoning RAG system with agentic workflows, tool calling, stateful memory and high-accuracy responses.

Kernl: The Operational Compiler for AI Agents

Kernl is a compilation layer that extracts messy operational judgment from Slack threads, tickets, and SOPs, compiling it into versioned, evidence-linked, executable skills files that any AI agent can consume to act like a company's best employee.

NeuroDesk AI Agent

NeuroDesk is a multi-agent AI system powered by AMD MI300X GPUs that autonomously executes full digital marketing campaigns — from audience segmentation and creative generation to real-time optimization — replacing hours of manual work in minutes.

System Aware Trader

S.A.T. (Sentinel-Aware Trader) solves the "hardware-blind" problem in High-Frequency Trading. Built for the AMD MI300X, it features a dual-layer architecture:​

CodeSentry

CodeSentry is an agentic AI security copilot that analyzes codebases, detects vulnerabilities, generates AI-powered fixes, streams AMD MI300X GPU metrics, and provides CUDA → ROCm migration guidance with Zero Data Retention.

ROCm Dev Assistant

We built an AI assistant that makes AMD GPUs easier to use. Using AMD MI300X hardware, we fine-tuned Llama 3.1 8B on ROCm/HIP docs to convert CUDA→HIP, debug ROCm errors, answer AMD GPU questions accurately through a simple chat interface powered by vLLM.

ThinkTank

ThinkTank: An AI platform where a Scholar Agent reads Rizal's novels and an Economist Agent reads live labor data — debating modern Philippine crises through history. Powered by Qwen3.5-27B-FP8 on AMD MI300X. Output: A Sukdulan verdict and policy essay.

OpsScout - Autonomous agent of event-driven SMBs

Turns "tourists are stuck" into action. 5 CrewAI agents on Qwen3 + AMD MI300X fuse weather, local events, and Airbnb occupancy into staffing, inventory, and comms proposals for tour operators and cafés which requires human-in-the-loop approval.

AgroFamiliApp

AgroFamiliApp: AI conversational agent (voice/text) for Brazilian family farmers, powered by AMD MI300X + Llama 3.1. Access rural credit, weather, markets, and agroecological technical assistance via WhatsApp, Telegram, or WebApp.

MediMesh — AI Triage for Rural Clinics

MediMesh is a 3-agent AI system for rural Indian PHC nurses. A Triager, DDx Reasoner, and Drug Safety Auditor run over Llama 3.2 1B — grounded in WHO IMCI guidelines — delivering a triage card in 60 seconds.

Sigui: The Trust Layer for AI Agents

Sigui: Vision-AI Security for agents. Powered by AMD MI300X/ROCm, it uses fine-tuned Qwen2-VL-7B to detect attack topologies in <50ms. It identifies threat "Topological DNA" that text engines miss. Real-time. Autonomous. Immutable. EIP-8259.

ProfiloAI - AMD GPU Doctor

ProfiloAI turns AMD GPU profiler output into a plain-English diagnosis, root cause, code fix, and expected speedup for ROCm developers.

PantryChefAi

PantryChef AI uses Llama 3.2 Vision on AMD MI300X GPUs to instantly scan groceries. It integrates USDA data for precise expiry tracking and Llama 3.2 11B to generate zero-waste recipes from expiring ingredients. Stop wasting, start tasting!

TactIntentNet

Decode latent tactical intent from broadcast positions using causal GNNs, detect opponent deviations in real-time, and generate LLM coaching alerts all on AMD MI300X.

Autonomous Enterprise AI Auditor

Multi-agent AI system that autonomously audits enterprise AI systems for compliance, bias, and performance risks using AMD GPUs and Qwen models. Delivers real-time risk scoring and actionable remediation recommendations.

ATOMiK Constitution Review

Git-native governance for coding agents: Rust/Git records, Lean judges, protected refs stay protected, and the model layer is wired for MI300X vLLM inference.

Financial Rule Trader TITAN

Neuro-symbolic trading signals: Qwen2.5-32B fine-tuned on AMD MI300X generates BUY/HOLD/SELL signals, each backed by a human-readable symbolic rule — not just a number. 91.6% accuracy, 96.5% on low-uncertainty signals.

MI300X Launch Doctor

AI deployment readiness agent that scans AI repositories for AMD ROCm risks, generates deployment fixes, and prepares Qwen3.5 MI300X/vLLM validation reports.

Vision Crafters

The AMD Multimodal Workbench is a unified app featuring three specialized modes: industrial defect inspection, an educational medical imaging workflow, and a vision-language assistant. It leverages cutting-edge multimodal models on AMD ROCm. 4

SpecSmith

SpecSmith is an agentic QA system that turns product specs, PRDs, and OpenAPI docs into risk-ranked test plans, executable test drafts, and coverage reports.

Autopsy investigates why companies fail

Autopsy investigates why companies fail — and what could have saved them. Six specialized AI agents per mode research in parallel, debate each other's findings, and produce a forensic verdict in ~22 seconds.

AI Signal, quiet struggles deserve loudest support

An AI powered mental wellness companion that helps users recognize early emotional struggles through supportive conversations and behavioral pattern awareness before conditions become severe.

Kitsune Engine: The Agent-Native Browser Engine

KitsuneEngine is a Rust-built browser agent engine that gives AI agents complete DOM sovereignty over the web, not screenshots, not CDP hacks, but full read/write access to live pages, secured by the compiler and gated by human-in-the-loop approval.

CatalystMD: AI-Powered Drug Discovery Platform

CatalystMD is an AI-powered drug discovery platform that helps scientists find promising drug candidates for difficult disease targets. It automates the complete drug discovery pipeline from protein structures to generating a scientific discovery briefs.

RoboControl

Voice-driven AI mission control for rover fleets: Qwen plans tasks, safety agents validate every move, and RoboControl routes commands to the right robot with telemetry-aware recovery.

Offsides

A multimodal AI system that analyzes UEFA Champions League match footage using YOLO + Qwen3-VL 32B on AMD MI300X to detect tactical signals that sports prediction markets misprice.

Project Hyperion: Adversarial Medical AI Swarm

Project Hyperion is an offline-first, adversarial multi-agent AI system delivering radiologist-level diagnostic triage. Powered by AMD MI300X, it features a 3-node consensus engine ensuring 100% data privacy and verifiable clinical accuracy at the edge.

ROCM-QLORA

Pure PyTorch QLoRA for AMD GPUs. No bitsandbytes, no CUDA. NF4/INT8 quant, Triton kernels, FP8 training, FSDP multi-GPU, SFT/DPO/GRPO trainers, HF plugin, GGUF/vLLM export. 117 tests. Runs on RX 7900 XTX and MI300X.

ROCm-Sentinel: Performance & Security Copilot

An intelligent agentic copilot that automates software migration to AMD ROCm via digital signature analysis, prioritizing real-time performance optimization and code security.

Security Council AI Agents and Fine-Tuned Models

AI agent orchestrator (Auditor and Builder) and fine-tuned Qwen2.5-14B models for automated code vulnerability detection & secure patch generation, optimized for AMD ROCm 7.0 + MI300X.

CreditBridge

From street to bankable — AI credit intelligence for African SMEs

TITAN-ROCm

TITAN-ROCm is an AI-powered GPU optimization system that analyzes PyTorch CUDA code, generates optimized variants using an LLM, benchmarks performance in real time, and selects the best implementation through autonomous optimization and profiling.

SentinelROCM

Privacy-first, multi-agent AI code analysis framework that runs entirely locally on AMD ROCm. No cloud, no API keys — your code never leaves the machine.

CORE AI: Multi-Agent ESG Evaluator for Mining

A multi-agent due diligence platform simulating a team of mining experts (Geologist, Engineer, Economist, COO). C.O.R.E. AI autonomously evaluates ESG risks, analyzes financials, and outputs institutional-grade investment decisions in seconds.

Code-Flow AI

An AI-powered tool that visualizes backend logic into interactive flowcharts and performs real-time security audits to detect vulnerabilities like hardcoded secrets and broken authentication in FastAPI/Python applications.

ROCmPilot

Fine-tuned Qwen2.5-Coder LoRA assistant that helps developers migrate CUDA-first AI workloads to AMD ROCm by detecting NVIDIA assumptions and suggesting ROCm-compatible fixes, commands, and debugging steps.

NeuralMix

Two-stage AI audio engineer: Kimi-Audio hears your stems, a fine-tuned model trained on AMD MI300X prescribes exact professional FX parameters — specific dB, Hz, ms values, not vague advice.

NeuroForge - Autonomous Software Engineer

Multi-agent system using AMD MI300X + Qwen2.5-Coder with planning, coding, review, and execution

Eidolon

Hypernetwork that generates task-specific LoRA adapters on-the-fly from few-shot examples, enabling instant domain specialization of LLMs without per-task fine-tuning. Built natively on AMD ROCm/MI300X

Aushadhi vault

Aushadhi Vault is an AI-powered pharmacy platform for Indian medical stores. It scans handwritten prescriptions using Llama 3.2 Vision on AMD MI300X, auto-generates bills, and monitors inventory with a smart agent that flags expiry and low stock.

RedCore Sentinel

A self-evolving AI security agent built on AMD MI300X. Three fine-tuned Qwen models detect, analyze, and patch adversarial attacks against LLMs in real time — and continuously retrain themselves on every threat they catch.

HexySAR: an AI-powered hexapod for SAR missions

HexySAR is an AI-powered autonomous hexapod for cave search-and-rescue. It explores hazardous terrain, detects survivors through vision and audio, maps safer paths, and sends rescue intelligence before human teams enter.

AsBuilt Lens: Autonomous Visual QA

A zero-shot multi-agent inspection system for manufacturing. It uses Qwen3-VL on the AMD MI300X to analyze components via live camera, and if it detects an error, a second agent automatically diagnoses the root cause. No model training required.

AiJockey — Multimodal AI DJ

AiJockey turns 3-8 of your audio clips into a broadcast-quality DJ mix. A multimodal Qwen2-Audio Director hears each track, plans the set narrative, and a 5-agent pipeline (Planner → Executor → Probes → Improver) renders it on AMD MI300X.

SachaSmart Shop AI Shopping Agent

An AI shopping agent that recreates a real store experience online and helps users choose the right products and sizes, reducing returns and improving trust in e commerce.

LIA — Autonomous AI with Evolving Identity

A modular autonomous AI agent inspired by the human brain. It features specialized LLMs for communication, coding, vision, and memory, with a Sims-inspired autonomy system of evolving traits, desires, dreams, and moods — running on AMD MI300X via ROCm.

Aubm AI Agent Orchestration

Aubm is an AI agent orchestration platform that turns a business goal into structured tasks, assigns agents, tracks execution, reviews outputs, consolidates evidence, and generates final reports.

AURA-DSP Autonomous Multi-Agent Audio Restoration

AURA-DSP is an autonomous multi-agent AI pipeline for digital signal processing. Utilizing AMD ROCm and vLLM, it intelligently separates, analyzes, and enhances audio tracks, optimizing complex engineering tasks through automated agentic workflows.

3D Engine for your Humanoid AI Agent

3D Engine for your Humanoid AI Agent. Integrated with Lipsync and a chatbox with file handling. A character controller. To make simulate how your AI agent interacts with the real world.

Vibe.run - The Strava for Developers

The Strava for developers. Every Claude Code session is silently captured by hooks, analyzed by Qwen2.5-Coder-32B on AMD MI300X, and turned into a shareable card. Code, ship, flex. See and share all your fellow developers AI session statistics.

Riprap: Agentic Flood Risk Intelligence for NYC

Riprap is a multi-agent civic AI generating cited flood briefings in 2 mins. Powered by IBM Granite, AMD MI300X fine-tunes, and Mellea rejection sampling, it replaces hours of manual engineering research with verifiable, audit-grade intelligence.

Lumi, AI voice helper for elderly with dementia

Lumi is a domain fine-tuned AI voice companion for dementia care, built on AMD MI300X. Persistent memory, voice-native responses under 1.5s, and scam deflection — fully local, no third-party APIs. Ranked 7th on EQ-Bench 3 out of 46 models.

AXIOM: The Autonomous SRE Agent

AXIOM is an autonomous SRE agent running on AMD MI300X that triages production incidents, executes code fixes, and opens GitHub PRs in under 90 seconds.

Khaleeji AI: World's 1st Bilingual Finance Agent

A specialized 7B parameter bilingual (Arabic/English) AI agent fine-tuned on AMD MI210 GPUs for the Gulf financial sector. It provides autonomous, real-time reasoning for UAE Tax, Saudi FDI, and Islamic Finance.

AI-Driven Industrial Control Agent

An intelligent mechatronic system leveraging the AMD Developer Cloud to provide real-time predictive maintenance and autonomous hardware control for industrial pumps.

Boardroom

Four specialist AI agents — Analyst, Skeptic, Strategist, and Auditor — deliberate over any uploaded document and deliver a cited, verified verdict. Built on Llama 3.1 70B running on AMD Instinct MI300X with vLLM prefix caching.

TradeGuard AI

Autonomous multi-agent system that audits trading strategies for risk exposure and FTMO prop firm compliance, powered by Qwen 2.5 72B running on AMD Instinct MI300X via ROCm-optimized vLLM and orchestrated through LangGraph.

PRISM: Optimizing Scalable Agentic Workflows

PRISM helps AI agents retrieve only the tools they need before acting, reducing prompt bloat, duplicate calls, tool confusion, and latency in scalable agentic workflows.

Qwen Medical

An smart medical assistant for both pacients and medical professionals

ContentDNA — Creative Fingerprint Engine

Upload any image, video, or audio and get your unique creative DNA — mood profile, style fingerprint, pacing analysis, audience prediction, and cultural references. Multimodal AI that reveals the hidden identity of any creative content.

Pheromone

Pheromone is a 5-agent AI recall OS for grocers. It traces affected products through supply chains, drafts 6-tier customer notifications, and adds first-of-its-kind Reassurance Notifications for unaffected buyers. Runs Qwen3-32B on AMD MI300X.

DeepSeek-Cybersec-33B: Autonomous Forensic Engine

Autonomous forensic engine based on fine-tuned DeepSeek-33B. Specialized in malicious intent detection, code de-obfuscation, and high-confidence threat assessment. Powered by AMD Instinct™ MI300X for real-time security triage.

QATALYST SDS — AMD Edge Demo

QATALYST SDS is an auditable agent orchestration framework running locally on AMD hardware. By exposing AI decision-making via a transparent four-stage loop, it ensures reliable, inspectable automation with zero cloud dependencies.

AMD-Optimized Agentic Workflow Engine

A high-performance AI agent framework built to leverage AMD ROCm and Instinct GPUs for local, low-latency execution of complex reasoning tasks and automated tool-use.

KiranaAI — Visual Billing Counter

An offline-first AI billing system for kirana stores that identifies products from photos using Gemma 4 + SigLIP and automatically adds them to the bill without barcodes or retraining.

Doc AURA

Doc AURA turns your codebase into living documentation, using agentic AI to understand changes and keep docs continuously up to date.

SignBridge — fine-tuned Qwen3-VL on AMD MI300X

Two people who couldn't communicate, now can. Real-time ASL (American Sign Language) → English speech, powered by our fine-tuned Qwen3-VL on AMD MI300X.

PolyM.AI.rket

Ask any binary question. A panel of AI analysts debates it in two rounds using live web research, then delivers a calibrated probability with a written synthesis.

DC-Ops-Assistant

DC-Ops teaches a 7B model to run a datacenter via physics-grounded RL. A reasoning teacher distills SFT data, then GRPO trains against a live RC thermal and power simulation. Built on Meta OpenEnv, NL command interface, runs on a single AMD MI300X.

UAE Expat Navigator

AI assistant that answers UAE government services questions using RAG over official RTA, ICP, MOHRE and GDRFA documents, with tool calling for eligibility checks and fee calculations.

Deep Chaos Scheduler with kernel optimization

A lottery scheduler for transformer fine-tuning — randomly selects 30–70% of layers, heads, and MLP channels each sticky block. Custom kernel optimizer cuts VRAM and trains 2.25× faster. Every seed beats full fine-tune on math evals.

Blitz SOC

Blitz SOC is an explainable hybrid SOC workflow combining deterministic governance with AI-assisted incident reasoning using CrewAI, Qwen, and AMD ROCm/vLLM inference.

AtlasOps

AtlasOps coordinates four specialist agents against live Kubernetes incidents—with real observability tools, approvals, Chaos Mesh scenarios, and training on AMD MI300X.

SOCrates

SOCrates is a multi-agent autonomous SOC analyst. Drop in a CVE — three AI agents fetch live threat intel, simulate the full attack path, and produce a MITRE ATT&CK–mapped IR report with a validated Sigma rule. Runs on AMD MI300X. Zero API fees.

KAAL — Knowledge Agent Arbitration Layer

4 adversarial AI agents debate every query to deliver calibrated sharp crisp ~50-year strategic forecasts. Fine-tuned Qwen2.5-7B on AMD MI300X with ROCm 7.0. Training loss: 2.5→0.47. Deployed via GGUF Q4 on HF.

TA Terminal

Full-stack equity intelligence terminal. Including live charts, financial information such as income and earnings reports, exotic option pricing, six-state market-regime classification, and FinBERT news sentiment, all accelerated by AMD Instinct MI300X.

LiteratiAI

LiteratiAI is an open-source, agentic writing assistant. Powered by LangGraph and vector memory, it dynamically tracks characters, plots, and lore across multiple novels to eliminate continuity errors.

Gemma 4 finetuned CUDA to ROCm kernel translator

Finetuning a Gemma 4 model to act as a CUDA-to-ROCm Kernel Translator creates a specialized tool capable of deep architectural reasoning between NVIDIA and AMD hardware.

SystemForge AI

SystemForge AI is a multi-agent AI platform that transforms outdated manual enterprise workflows into production-grade AI-native systems using Architect, Critic, Refiner, and Executive agents.

NourAI : Your AI Recovery Companion

A multi-agent AI companion for eating disorder recovery. Delivers a personalized daily card — mood-matched meals, art therapy, color palettes, and progress tracking. Built on AMD GPU with Qwen and FLUX diffusion.

DocuMind AI — Multi-Modal Document Intelligence

Upload any document. 4 AI agents on AMD MI300X GPUs via Fireworks AI analyze it — Vision Agent (Kimi K2.5), Reader, Analyst, Reporter — and deliver a structured intelligence report in under 60 seconds.

ROCmBench Agent

ROCmBench Agent auto-loads Hugging Face models on AMD GPUs with vLLM + ROCm, runs concurrency benchmarks, checks GPU evidence, and recommends the best production deployment config.

EarningsPilot AMD

EarningsPilot AMD automates financial research using a multi-agent system KPI, Risk, Thesis, Report to synthesize SEC filings and transcripts. Accelerated by AMD Instinct MI300X, it delivers source-grounded, structured investment intelligence in seconds.

Chaos Economy: Multi-Agent RL Market Simulator

Multi-agent RL sim: 4 traders, market maker & SEC regulator trained via GRPO on Llama-3.2. 250 steps of emergent financial crisis — slaughter, adaptation, collusion, regulatory oversight — no scripts, just learned behavior.

PakFreelance: AI Freelance Toolkit

AI-powered toolkit for Pakistani freelancers using 5-agent CrewAI, Llama 3.3 70B on AMD MI300X, and Streamlit. Generate proposals, detect scams, calculate rates instantly.

CLOAK-AI

An autonomous AI privacy engine powered by AMD Instinct™ MI300X and Llama-3.3-70B that detects and masks sensitive PII/PCI data in real-time with 0.08s latency.

Octal 007: Code -Domain Invariant Hunter On AMD

Fine‑tuned Qwen2.5‑Coder‑32B on MI300X to discover periodic invariants in impartial games & generate Lean‑4 conjectures – first open‑source ROCm‑native pipeline. 🚀

Agent I C - A true autonomous RTL to GDSII Agent

My agent goes from Prompt to GDSII , by doing all the vlsi checks and corrects them using feedback loop . It is based on crewAI orchestration . It supports all the opensource PDKs.

Combo - A Multi-Model AI Chat

Combo is a locally-run AI chat system that queries three different language models back-to-back on the same prompt, then synthesizes their answers into a single unified response.

FORGE — Autonomous Compute Exchange

FORGE is an agentic compute exchange that cuts AMD GPU costs by batching jobs, timing spot-price dips, and subleasing idle instances — settling every transaction with X402 micropayments and streaming live agent decisions to a real-time dashboard.

ClimateGuard

ClimateGuard is an AI-driven platform for climate risk intelligence that provides real-time weather insights, disaster warnings, and individualized safety advice through Qwen 2.5, which is based on AMD MI300X GPU with ROCm and Ollama.

SkillForge — AI Job Navigator

AI-powered career intelligence platform using Qwen3 and live job market data to match users with relevant careers, identify skill gaps, analyze demand trends, and generate personalized learning roadmaps.

Insight - Real Time Copilot for the OR

Insight is an AI co-pilot for the operating room. Insight continuously ingests live video and voice to auto-generate operative notes, flag technical errors in real time, and surface intra-op decision support.

AgentOS: AMD-Powered Multi-Agent Workflow Console

AgentOS is a multi-agent AI workflow console that runs planning, execution, validation, and memory recall using Mistral 7B served through vLLM on AMD GPU cloud infrastructure.

Why Agent: Autonomous Root Cause AI Agent

why-agent is an enterprised level autonomous root-cause intelligence tool for data teams. It investigates your dataset, forms and tests hypotheses, and returns a structured report with evidence chain, quantified attribution, and next steps.

HIP-Hop: CUDA to AMD Migration Swarm

HIP-Hop is a multi-agent AI swarm that autonomously translates legacy NVIDIA CUDA code into AMD ROCm/HIP C++ syntax, eliminating vendor lock-in using local MI300X compute.

AgentFreeFood

Discover free food near you , after spending all your money on claude code

NexusCore Systems

Building high-performance, automated digital solutions with a focus on scalable Node.js architecture and futuristic user interfaces

Resep ID Gemma 4 E2B IT - Indonesian Recipes Model

Fine-tuned Gemma 4 on Indonesian recipes. Unlike the base model, it generates the right recipe for the right dish, not just generic Indonesian food.

School Data Validation and Education Intelligence

EduGuard AI is an AI agent-driven system that automatically validates school education data, detects anomalies, and generates actionable insights for education planning using AMD GPU-powered inference and open-source large language models.

Ghost-Coder: Autonomous CUDA-to-HIP Agent

Ghost-Coder is an autonomous AI agent powered by Qwen2.5-Coder-32B and AMD MI300X that dynamically translates, verifies, and self-heals CUDA kernels into AMD ROCm/HIP code, accelerating cross-platform GPU migration.

Kazi AI HR Assistant

Kazi is an AI-powered HR assistant that helps employees instantly access company policies, leave balances, and payroll information securely using conversational AI, intelligent document search, and role-based access control.

Agentra

Agentra is an agentic AI-powered content creation system that uses multiple AI agents to plan strategies, research trends, generate content, optimize engagement, and improve workflows through feedback and memory.

Avocado Orchard Digital Twin AI Application

AI-powered avocado orchard command center using Cesium, AMD MI300X inference, orchard detection, PostgreSQL archives, and 3D digital twins to scan regions, map orchard parcels, analyze risk, and guide grower decisions.

NCERT Based Local LLM For Low end devices

This Model is a GRPO-RL Based Fine-tuned version of Llama-3.2-3B. It utilizes GRPO And NCERT Dataset to Produce Perfect Replies to NCERT Based Queries for Secondary, Senior Secondary and High School Learning. 4 Bit quantized to run even on Low End devices

ASAR - Autonomous Search And Rescue system

An agentic Search And Rescue system with full vision and sensing capability, embodied in a physically-accurate simulation environment.

Neural-Ink: Real-Time Stylized Universes

A high-performance Pervasive AI pipeline that transforms complex 3D environments into stylized hand-drawn art in real-time, powered by AMD Instinct™ MI300X and optimized for local execution on AMD Radeon™ hardware.

MachinaCheck

AI powered multi-Agent app that uses AI to analyze that the give component could be manufacture or not on the basis of shop inventory

AEGIS Lab

AEGIS Lab: Computational physics platform exploring whether quantum vacuum forces in metamaterials can couple to gravity. Validated Casimir theory. Testable experiment proposed.

Qubitpage Sentinel Vision Navigator Android APP

Sentinel Vision Navigator - Android APP, is an assistive AI system designed to act as a practical pair of eyes for blind and partially sighted users.

VaaniGuru

AI voice tutor for 300M Indian students — speak in Hindi, Bengali, Tamil or 10 other Indian languages, get correct answers back in your language with voice. Qwen2.5-7B fine-tuned on AMD MI300X.

SafeSite AI

SafeSite AI exists to act as a real-time safety officer for every CCTV in construction sites; it catches violations instantly, calls the worker out by name, logs the violations and stops the next accident before it ever happens.

SAGE IDE — Adversarial Multi-Agent Code Engine

Four AI agents debate your code in real time. SAGE IDE's adversarial boardroom — Architect, Implementer, Synthesizer, Red-Team — resolves via Nash Crucible on AMD MI300X ROCm. The friction is the feature: +12.4% HumanEval+, +8.2% SWE-Bench-Lite.

ROCKIT - Enable AMD Hardware

ROCKIT is an AMD-native agentic AI platform — the missing software layer that makes AMD hardware as deployable as NVIDIA's. One SDK, zero fragmentation, batteries included: inference, RAG, multimodal vision, and orchestration — all on AMD silicon.

Synthex — Autonomous Research Analyst

RAG retrieves. Synthex investigates. A 6-agent pipeline that plans, searches, detects contradictions, and self-corrects — then exposes the full research pipeline as an MCP server so any AI agent can call it as a tool.

GhostWriter OCR: AI-Powered

An AI-powered document intelligence tool that transforms handwriting into actionable data. It interprets intent, detects errors, and uses agentic workflows to automate form processing—all optimized for high-performance execution on AMD hardware

Majlis the AI models discussion rooms

An app where you chat with multiple models while deciding on whom to stay and whom to kick with the help of an AI curator learning on the fly.

NavSight AI: Navigation for the Visually Impaired

Real-time AI navigation for the visually impaired using 4 specialized agents on AMD MI300X. Detects objects, estimates depth, understands scenes via VLM, and delivers spoken hazard alerts with priority ranking.

ROCmVision: AI Traffic Violation Detection

An AI-powered traffic violation detection system using custom YOLO models, ANPR detection, and AMD ROCm acceleration on MI300X GPUs for real-time helmet, rider, and license plate analysis.

Health Passport AMD

Health Passport AMD turns prescriptions, lab reports, and medical receipts into a patient-owned longitudinal health record using multimodal models on AMD cloud GPUs, then reconciles medications, follow-ups, and timeline entries automatically.

Auto Learning

An autonomous pipeline that teaches a language model to become a domain expert by reading research papers, extracting verified knowledge, resolving contradictions, and continuously fine-tuning itself, compounding intelligence with every cycle.

CareerForge AI

AI-powered career prep platform: search live jobs, parse & tailor resumes to postings, and run voice mock interviews with LLM feedback — all locally via Ollama, with no database or accounts required.

TicketMind AI

An agentic IT ticketing system that analyzes problems in plain English, auto-resolves known issues from a knowledge base, and intelligently assigns new tickets to the best available expert using Qwen2 on AMD ROCm.

LLM-Diff

LLM Diff is an open-source evaluation engine that runs a parallel behavioral battery against two models and outputs a Behavioral Fingerprint Distance. It probes 6 dimensions: Sycophancy, Refusal, Hallucination, Calibration, Reasoning, and Verbosity.

MediVision: Dermatology & Wound Care AI Assistant

MediVision is a multilingual AI assistant for dermatology and wound care that analyzes skin condition images combined with patient symptom descriptions, powered by Qwen2.5-VL-7B running on AMD MI300X via vLLM.

SwarmAudit

SwarmAudit is a multi agent code audit dashboard that scans GitHub repositories for security, reliability, dependency, observability, performance, and production readiness risks.

AppSec Hunter: Llama 3.1 70B Scanner

An automated zero-day vulnerability hunting pipeline that uses an ephemeral AMD MI300X backend to slash compute costs while analyzing raw HTTP traffic with Llama 3.1 70B

Learnable: Adaptive Study via Error Reinforcement

Learnable transforms any syllabus into an interactive knowledge graph. AI tutor chats per topic, tracks your mistakes, generates flashcards and notes, and prepares you for deadlines — all powered by AMD ROCm and agentic LLM workflows.

Vortex Compressor Float32 Compression

It is a lossless neural compressor for high energy physics data. Using a novel float32-aware embedding (TDTEmbedding) and trained on an AMD MI300X GPU, it improves compression by up to 28% over LZMA-9 on ATLAS datasets.

Med Audit Swarm

Med-Audit-Swarm is an autonomous AI swarm powered by AMD MI300X that prevents fatal prescription errors. It extracts messy EMR data to detect lethal drug interactions with 100% precision, eliminating alert fatigue with zero hallucinations.

OmniDoc — Talk to Any Document

Upload complex documents — PDFs with charts, tables, scans — and have a natural languageconversation about everything, including the visuals.

Centinela Agent

An AI agent that turns field photos from messaging groups from security technicians into structured compliance reports — eliminating 12+ hours of manual work per site, powered by Gemma 4 31B on AMD Instinct MI300X

ROCm AgentOps

An AgentOps command center that scores, routes, audits, and optimizes critical AI workflows using deterministic controls, Qwen 7B on AMD/vLLM, live workload signals, policy guardrails, and human escalation.

ROCm Studio

A guided web app for cleaning documents, fine-tuning Gemma with LoRA on AMD MI300X GPUs, and deploying a Try-It inference endpoint without touching a terminal. For non-technical users who want to fine-tune a small, local model, this does it all for them.

Boundary Forge

Boundary Forge helps enterprises safely deploy any LLM by automatically finding model failure boundaries, benchmarking risk, and converting failures into runtime guardrails. Demonstrated with Qwen 2.5-72B on AMD Cloud.

OrdoFlow — AI Quote-to-Cash Copilot (AMD Track 1)

AI Quote-to-Cash Copilot turns quote context into prioritized next actions, risk flags, and customer-ready messaging inside OrdoFlow’s live quote workflow.

Thor v2 — RAG-Free Fitness Intelligence

Thor v2 fine-tunes Qwen3-8B on AMD MI300X to answer fitness questions from weights alone. No RAG. Citation keys are emitted inline and validated against a locked registry. 7,118 training examples. 14 population profiles. 100% JSON contract pass rate.

Agronomist

Resume Compact

An intelligent HR analysis system powered by FastAPI and local LLMs that automates resume screening and candidate evaluation.

AI Resume Intelligence System

AI-powered resume processing and job matching system using vector embeddings, semantic search, and AMD-backed Fireworks API LLMs for intelligent candidate analysis and job recommendations.

GeoRescue: Omni GIS Agent

An AI-powered disaster response platform that combines satellite vision, live flood intelligence, and multi-agent GIS reasoning to detect flood impact, identify blocked roads, and generate the safest evacuation route in real time.

Swarm

A multi-agent collective intelligence for DevOps orchestration and automation, powered by the GPUs of the AMD Developer Cloud.

AI-ERP Integration

An AI-driven bridge that automates the business lifecycle by integrating AI agents with ERPNext. It handles lead qualification, project management, and automated invoicing, using AI Vision for seamless payment reconciliation.

TempoGraph: Local Multimodal Video Analysis

A fully-local multimodal video analysis pipeline that transforms raw video into structured entities, events, and timelines using YOLO26, Whisper.cpp, Depth Anything V2, and Qwen3.5-VL — all running on consumer AMD hardware. No cloud, no API keys.

GPU Cost Doctor

GPU Cost Doctor helps AI teams diagnose GPU inference cost, deployment risk, fine-tuning readiness, and benchmark gaps before moving models to production.