Anthropic Anthropic Claude Applications

CogniRoute

Drupal AI Factchecker — Claim-Level Truth on AMD

Native Drupal module that fact-checks claims against a university's own corpus, scores AI-writing likelihood, and finds verbatim plagiarism — all inference running on local AMD hardware (ROCm).

RouteZero: Hybrid Token-Efficient AI Agent

A hyper-optimized routing agent that guarantees high accuracy while crushing token costs. It uses strict deterministic verification and a local-first cascade to answer questions for zero tokens, only escalating to Fireworks when mathematically necessary.

Trippy Troppy

Token-minimal agent for the Fireworks AI track. Classify each task locally with regex, then make exactly one call per task with a terse category-specific prompt and the cheapest suitable allowed model.

AI SOC Analyst

SentinelAI SOC is a multi-agent AI platform that autonomously analyzes security logs, detects cyber threats, maps attacks to MITRE ATT&CK, generates Sigma/YARA detection rules, enriches incidents with threat intelligence, and produces incident reports.

GenRisk

A multi-omic cardiometabolic risk reader that layers genomic, methylation, microbiome, and clinical evidence side by side — every number traceable to a published source, never fused into a single misleading score.

CaprIA – Smart Goat Pattern Identification

AI system that identifies goat breed patterns from head photos. Via chatbot, the user sends an image and receives the identified pattern, based on features such as horns, ears, and nasal profile.

Cascade Router: Local-First AI Agent

A cascade-routing AI agent that answers locally first at zero token cost and escalates to Fireworks AI only when the local answer can't be trusted, across all 8 required task categories.

Grounded Multi-Style Video Captioning Agent

A Dockerized Track 2 video-captioning agent that samples video frames with ffmpeg, uses Fireworks/Kimi vision inference, and generates grounded captions in formal, sarcastic, humorous-tech, and humorous-non-tech styles.

Crucible — the autonomous AI underwriter

The autonomous AI underwriter that puts loans on trial, works the conditions, and rebuilds denials — every claim cited to real guideline text, running on AMD Instinct MI300X.

RouteZero — Token-Efficient Task Agent

A Track 1 routing agent that completes all eight task categories while spending the fewest tokens possible, deciding per task, in real time, how little inference each answer actually needs. :) :-)

Agentic EWS Advisory system

the app will help Smallholder farmers in agricultural zonesget seasonal forecast and advisories to reduce planting-window errors, crop failure, and increase uptake of insurance cover owing to emergence of climatic regimes. advisory tools

ChiefFlow AI — AI Chief of Staff

An AI Chief of Staff that turns emails, documents, and meeting notes into automated, human-approved workflows, powered by 6 specialist AI agents.

ArchMind: AI Software Architect

ArchMind transforms any codebase into an intelligent knowledge graph that AI agents can reason over. Developers can explore architecture, analyze dependencies, assess change impact, detect design issues, and generate documentation using natural language.

VoxRouter — Hybrid Token-Efficient Routing Agent

An AI router that sends each task to the cheapest model that can handle it — free local models via Ollama for simple tasks, remote models only when complexity demands it, with confidence-based escalation and a built-in benchmark score.

StyleFrame Caption Agent

A containerized video-captioning agent that samples video frames, creates a contact sheet, and uses Fireworks multimodal models to generate accurate formal, sarcastic, humorous-tech, and humorous-non-tech captions.

Axiom - Aether

Axiom is a local-first Rust research runtime for online Test-Time Training (TTT), context compression, grounding checks, and self-healing execution loops.

Token Sage Application

TokenSage is a hybrid AI routing system that intelligently directs simple queries to a local AMD Ryzen AI NPU and complex queries to cloud LLMs, reducing token usage, lowering costs, improving response speed, and enhancing user privacy.

FairPrice: AI Agents for Africa's Informal Economy

An AI-powered escrow marketplace solving cash-on-delivery risk for Africa's informal sellers; autonomous agents (ZEMA 360) run fulfillment and pricing on Qwen 3.7 Plus, served on AMD Instinct GPUs via Fireworks AI, live in production today.

SellerKavach

SellerKavach is an AI order intelligence layer for chat-based sellers. It intercepts messy WhatsApp/IG chats, extracts structured orders, predicts RTO risk before shipping, and builds a shared buyer trust network to stop fraud and losses.

SatyaPath Verify: AI Welfare Claims Auditor

AI ensemble + SHAP + Gemma flag and explain fraudulent MGNREGA welfare claims in plain language, then cryptographically commit verified batches via Merkle proofs, built on AMD Developer Cloud and Fireworks AI.

FireCascade Agent

General-purpose AI agent for AMD Hackathon Track 1. Handles 8 NLP/coding task categories via a cost-optimal Fireworks model cascade-cheap models first, escalating only when needed to minimize token spend while passing the accuracy gate.

ProcureMind AI – Enterprise Procurement Agent

An AI-powered procurement assistant that analyzes Bills of Quantities (BOQs), estimates project costs, recommends vendors, optimizes procurement decisions, and generates commercial quotations through an intelligent conversational interface.

Mallana: AI Runtime for Autonomous Development

Mallama is an open-source AI runtime built on a simple belief: the future of AI shouldn't belong only to those with the biggest infrastructure. By making AI development more efficient, Mallama helps anyone build more with less.

Hermes v3 — The Divine Messenger of Task Routing

Hermes classifies each task with zero-cost heuristics, then routes it to a category-tuned model, prompt, token budget, and reasoning effort.

Video Captioning Agent

An agent that watches a video clip, samples frames with ffmpeg, and generates faithful captions in four required styles: formal, sarcastic, humorous tech, and humorous non-tech.

Kizuki: Multi-Provider LLM Agent Pipeline

Kizuki is a smart LLM agent pipeline that classifies tasks into 8 categories, routes each to the optimal model tier, and generates answers using any of 100+ LLM providers via LiteLLM. Supports Fireworks AI and more. Includes CLI and web UI modes.

Multimodal Video Captioner

A multimodal AI pipeline that turns short videos into accurate captions with personality, combining visual understanding, speech analysis, OCR, and LLM reasoning across four distinct styles.

Memory Recall Assistant

A private, on-device memory assistant for cognitive decline. Using ExecuTorch on Snapdragon 8 Elite NPU, it identifies faces, transcribes conversations via Whisper, and generates warm recall cues via Llama 3.2 1B - fully offline, zero cloud.

OBSERVA

OBSERVA is an offline-first, on-device Android assistant for blind and low-vision users that provides real-time hazard-aware guidance, scene understanding and accessibility feedback using private edge AI without sending camera data to the cloud.