Reinforcement Learning Applications

AEGIS — Autonomous Enterprise Intelligence OS

AEGIS is an autonomous AI platform that continuously patroles the live web for threat vectors, GTM hiring signals, and pricing pivots. It scrapes, reasons, remembers, and dispatches automated webhooks giving teams always on intelligence.

FeatureFlag : Autonomous AI Decison Rollout System

FeatureFlag is an autonomous AI copilot for continuous deployment that uses reinforcement learning, anomaly detection, and multi-agent simulation to automate feature rollouts, instant rollbacks, and enterprise-scale release governance.

AI powered stock analysis platform

An AI-powered stock analysis platform featuring a real time live dashboard, sentiment driven news feed, and Prophet based price forecasting that gives investors a data driven edge in the market.

Anvaya EnterpriseIQ

Anvaya EnterpriseIQ: An AI-driven intelligence platform for large-scale data action. Features unified RAG chat for documents, NL-to-SQL analytics for massive CSVs, real-time anomaly detection, and interactive knowledge graphs to visualize complex relation

ARCHON: AI for Engineering Intelligence

ARCHON is an AI engineering intelligence platform powered by IBM Bob that helps developers understand codebases, analyze architecture, predict code impact, debug faster, and manage complex software systems using repository-wide AI reasoning.

Trident - Redefining AI security

TRIDENT is a 12-layer AI safety scanner that instantly analyzes any prompt or response for toxicity, jailbreaks, PII, bias, misinformation, injection attacks, and more — powered by a 74k-row ML model with real-time Gemini deep analysis.

GHOST Chimera

local-first agent orchestration runtime. it acts as the user proxy, AI agent that is trained on the user's identity, local machine, emails and converts them into trainable datasets. a personal AI proxy (ghost)

SOVEREIGN: Self-Evolving Enterprise AI Immune

Self-evolving AI security platform. Behavioral DNA detects 95% of attacks vs 40% rule-based. Honeypot capture, Gemini mutation, pre-deploy policies, retrain model, adversarial arena, cognitive layer. Built on Veea Lobster Trap + Gemini.

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.

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

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.

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.

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

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.

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.

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.

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.

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

Axiom

Axiom is a multi-agent AI pipeline automating academic systematic reviews. Built with LangGraph, vLLM, Qwen2.5-7B, and QwQ-32B on AMD MI300X, it executes PRISMA screening, data extraction, and gap analysis to generate APA-7 literature reviews.

ROCm Forge: 9-Agent CUDA to AMD Migration Engine

ROCm Forge: 9-agent architecture cuts down CUDA-to-ROCm migration effort by ~65% It replaces manual regex with AI precision; provides hardware-aware AST translation, risk heatmaps, build-error prediction and AMD Developer Cloud artefacts.

P402 Meter: Proof-of-Spend for AI Agents

P402 Meter turns every AI action into a priced, governed, and auditable USDC event on Arc, giving agents a payment trail, budget cap, and receipt for each unit of work.

Agent-to-Agent Compute Marketplace

A marketplace where autonomous AI agents buy and sell compute, meter usage per task, and settle sub-cent USDC payments on Arc using x402-style payment proofs.

AUGMENTIC

PYTHAI: The Augmentic Smart Agent Platform Architecting Humanity's Transition from Information Age to Knowledge Economy PYTHAI Augmentic Smart Agent Platform - infrastructure to transform raw information into distributed knowledge

InfernoBots

InfernoBots is an emergency response force with autonomous drone scout wildfires supporting real-time vision intelligence while ground robot(s) mobilize fire trucks and supplies. it aims to turns disaster response faster and higher precision.

AI-Powered Autonomous Fire Response Drone

AI-powered autonomous drones that detect fires, navigate complex environments, and deliver real-time actionable insight to emergency responders

FleetMind

FleetMind is a web-based "Digital Twin" for autonomous robot fleets. It uses Gemini 2.0 Flash for natural language command parsing and local Q-Learning agents for collision avoidance and battery optimization in a real time 3D simulation

Data infrastructure for Real World Robot Training

We built a data infrastructure platform to capture and convert real-world human demonstrations into reusable robot training data using XR devices.

FIS AEGIS

Financial Intelligence Swarm: FIS AEGIS: Multi-Agentic Anti-Fraud, FinOps & Compliance System

Quantified Ubuntu Contribution Integrity Crucible

QUBIC is the C++ Smart Contract DAO on Qubic that transforms life-saving clinical impact into immutable, weighted governance power. It eliminates bias and rewards contributions to Ubuntu Patient Care with transparent, dynamic merit-based credentials