Track 2, Script-Kitty01, ForgeAI by Script-Kitty01 · Pull Request #325 · AMD-DEV-CONTEST/Radeon-hackathon-2026-07

Script-Kitty01 · 2026-08-06T17:41:20Z

ForgeAI -- Your Private AI Engineering Team

A conversational multi-agent AI engineering assistant running 100% locally on AMD Radeon GPUs via ROCm/HIP.

#	Item	Included
1	Project Spec (PDF)	`PROJECT_SPEC.pdf`
2	Source Code	https://github.com/Script-Kitty01/devMaster_amd
3	Demo Video	https://drive.google.com/drive/folders/1PHfQT8CkQi6C3Jq6fY6C_3cnXVxyGfLl
4	PPT	`ForgeAI_PPT.pptx`

6-agent collaborative system: Planner, Security, Performance, Architecture, DevOps, Consensus
100% local on AMD Radeon: Llama 3.2 3B Instruct Q4_K_M via llama-cpp-python with HIP/ROCm
124 tok/s generation on gfx1100 (RDNA 3) with ROCm 7.2.1
15.5x faster than CPU-only inference
2.57 GB VRAM usage -- fits on consumer GPUs
LangGraph orchestration, ChromaDB RAG, Streamlit UI
Tools: Bandit, Semgrep, GitPython, Dockerfile Validator

- PROJECT_SPEC.pdf: Full project specification document - ForgeAI_PPT.pptx: 10-slide presentation - SUBMISSION_README.md: Submission overview with all links Source: https://github.com/Script-Kitty01/devMaster_amd Video: https://drive.google.com/drive/folders/1PHfQT8CkQi6C3Jq6fY6C_3cnXVxyGfLl