title: SynapseOS
emoji: 🧬
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 6.14.0
app_file: app.py
pinned: true
license: mit
🧬 SynapseOS — AI Agent Civilization
5 Expert AI Agents that Think, Debate, and Decide — Powered by AMD MI300X GPU
🎯 What is SynapseOS?
SynapseOS is a multi-agent AI debate system where 5 specialized AI agents independently analyze any business idea or problem — each bringing a distinct professional perspective — and collectively arrive at a GO / CONDITIONAL GO / NO-GO decision.
Built for the AMD Developer Hackathon 2026, SynapseOS runs Qwen2.5-0.5B-Instruct via vLLM on AMD MI300X GPU infrastructure, delivering fast, structured, and intelligent multi-agent reasoning.
Think of it as assembling a full expert boardroom — a Project Manager, Senior Developer, Devil's Advocate, Financial Analyst, and Security Expert — all debating your idea simultaneously in seconds.
🖥️ Live Demo
🌐 Space URL: https://huggingface.co/spaces/lablab-ai-amd-developer-hackathon/synapseos
🤖 The 5 Expert Agents
| # | Agent | Role | What It Delivers |
|---|---|---|---|
| 1 | 🧠 PM Agent | Project Manager | Phases, timeline, milestones, GO/NO-GO |
| 2 | 💻 Developer Agent | Senior Developer | Tech stack, architecture, scalability |
| 3 | 🔍 Critic Agent | Devil's Advocate | Risks, flaws, failure scenarios |
| 4 | 💰 Finance Agent | Financial Analyst | Costs, revenue model, break-even |
| 5 | 🔒 Security Agent | Security Expert | Vulnerabilities, GDPR, auth strategy |
Each agent receives the same idea but analyzes it through its own professional lens — producing 150+ word structured responses independently.
✨ Key Features
- ⚡ AMD MI300X Powered — vLLM inference server running on AMD GPU hardware
- 🤖 5 Parallel AI Agents — Each agent calls the model independently with unique system prompts
- 📊 Structured Analysis — Every agent delivers 5-point detailed breakdown
- 🎯 Final GO/NO-GO Decision — PM Agent synthesizes all perspectives into a verdict
- 🔊 Voice Summary — Full English text-to-speech audio summary via gTTS
- 🧠 Session Memory — Tracks and displays all ideas analyzed in the session
- 🌐 Public Share Link — Instantly shareable Gradio link
🏗️ Architecture
┌─────────────────────────────────────────────────────────┐
│ USER INPUT (Idea) │
└─────────────────────┬───────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────┐
│ SynapseOS Orchestrator (app.py) │
│ Gradio UI Interface │
└──┬──────┬──────┬──────┬──────┬──────────────────────────┘
│ │ │ │ │
▼ ▼ ▼ ▼ ▼
PM Dev Critic Finance Security
Agent Agent Agent Agent Agent
│ │ │ │ │
└──────┴──────┴──────┴──────┘
│
▼
┌──────────────────────────────┐
│ vLLM OpenAI-Compatible │
│ API Server (Port 8000) │
│ AMD MI300X GPU Instance │
└──────────────┬───────────────┘
│
▼
┌──────────────────────────────┐
│ Qwen2.5-0.5B-Instruct │
│ Running on ROCm / AMD GPU │
└──────────────────────────────┘
│
▼
┌──────────────────────────────┐
│ Final Decision + Voice │
│ Summary (gTTS) │
└──────────────────────────────┘
🛠️ Tech Stack
| Layer | Technology | Purpose |
|---|---|---|
| GPU Compute | AMD MI300X | High-performance AI inference |
| ML Framework | ROCm + vLLM | OpenAI-compatible inference server |
| AI Model | Qwen2.5-0.5B-Instruct | Fast, efficient language model |
| UI Framework | Gradio 4.44 | Web interface |
| API Client | OpenAI Python SDK | vLLM API communication |
| Voice | gTTS | Text-to-speech summary |
| Hosting | HuggingFace Spaces | Public deployment |
🚀 Local Setup
Prerequisites
- Python 3.11+
- HuggingFace account + API token
- AMD GPU with ROCm (for local vLLM) or AMD Developer Cloud access
1. Clone Repository
git clone https://github.com/exedistrict-ux/synapseos.git
cd synapseos
2. Create Virtual Environment
python -m venv .venv
# Windows
.venv\Scripts\activate
# Linux/Mac
source .venv/bin/activate
3. Install Dependencies
pip install -r requirements.txt
4. Configure Environment
Create .env file:
HF_TOKEN=hf_your_token_here
VLLM_BASE_URL=http://your_amd_gpu_ip:8000/v1
MODEL_NAME=Qwen/Qwen2.5-0.5B-Instruct
5. Start AMD vLLM Server (on AMD GPU instance)
pip install vllm
python -m vllm.entrypoints.openai.api_server \
--model Qwen/Qwen2.5-0.5B-Instruct \
--gpu-memory-utilization 0.3 \
--max-model-len 2048 \
--port 8000
6. Run SynapseOS
python app.py
Open: http://127.0.0.1:7860
🧪 Running Tests
python test.py
Expected output: ```
SynapseOS — Test Suite AMD Developer Hackathon 2026
[PASS] Environment Variables (.env) [PASS] Python Imports [PASS] HuggingFace InferenceClient [PASS] PM Agent API Response [PASS] All 5 Agents API Response [PASS] Text-to-Speech (gTTS) [PASS] Memory System [PASS] Gradio UI Components
============================================================ Results: 8/8 tests passed All tests passed! SynapseOS is ready.
---
## 📁 Project Structure
synapseos/ ├── app.py # Main application — 5 agents + Gradio UI ├── test.py # Full test suite ├── requirements.txt # Python dependencies ├── .env.example # Environment variable template ├── .gitignore # Git ignore rules └── README.md # This file
---
## 💡 Example Output
**Input Idea:** *"Build a scam protection app for senior citizens in India"*
PM Agent → GO ✅ — 3 phases, 6 month timeline, team of 4 Developer Agent → React Native + FastAPI + PostgreSQL + AWS Critic Agent → Market saturation risk, digital literacy gap Finance Agent → $45K dev cost, break-even at 800 users Security Agent → OWASP compliance, biometric auth required
Final Decision: CONDITIONAL GO 🟡 Action: Survey 100 seniors → Build MVP → Partner with NGOs Biggest Risk: Low smartphone adoption in target demographic
---
## 🏆 AMD Developer Hackathon 2026
**Event:** AMD Developer Hackathon — lablab.ai
**Dates:** May 4–10, 2026
**Prize Pool:** $21,500+ + AMD Radeon AI PRO R9700 GPU
**Track:** AI Agents & Intelligent Workflows
**Team:** Gaurang_Solo
### Why AMD?
- AMD MI300X delivers **192GB HBM3 memory** — ideal for LLM inference
- **ROCm** open-source stack enables flexible model deployment
- **vLLM on ROCm** provides OpenAI-compatible API with AMD GPU acceleration
- $100 AMD Developer Cloud credits enabled rapid prototyping
---
## 📄 License
MIT License — see [LICENSE](LICENSE) for details.
---
## 🙏 Acknowledgements
- [AMD Developer Cloud](https://www.amd.com/en/developer) — GPU infrastructure
- [vLLM](https://github.com/vllm-project/vllm) — High-throughput LLM serving
- [Qwen Team](https://huggingface.co/Qwen) — Qwen2.5 model family
- [Gradio](https://gradio.app) — UI framework
- [lablab.ai](https://lablab.ai) — Hackathon platform
---
*Built with ❤️ on AMD MI300X · ROCm · vLLM · Gradio*