An intelligent requirement extraction engine that listens to client meetings, handles pivots in real-time, and generates production-ready code.
DevDraft AI is a multi-modal agentic tool designed for high-stress hackathons and rapid prototyping. It acts as an AI Senior Product Manager that joins your meetings, understands context, and converts spoken requirements directly into code.
- Real-Time Pivot Handling: "Actually, let's use Supabase instead of Firebase." → DevDraft immediately updates the specs and strikes through the old requirement.
- Live PRD Generation: Generates a structured Product Requirement Document (PRD) on the fly.
- Instant Code Generation: One-click generation of a full React + Vite + Tailwind project.
- Agentic Tool Integration: Exports optimized prompts for Lovable, Bolt.new, and Replit Agent.
The project consists of three main components:
| Component | Location | Tech Stack | Description |
|---|---|---|---|
| Extension | /extension |
React 19, Vite, Chrome Manifest V3 | The side-panel companion that captures audio and displays the live feed. |
| Web Dashboard | /landing |
React 19, Tailwind, Framer Motion | The browser-based experience for users without the extension. |
| Cloud Backend | /backend |
Python 3.10, FastAPI, Gemini 1.5 Pro | The brain. Handles transcription (Deepgram), logic (Gemini), and code generation. |
You don't need to install anything to try DevDraft AI.
- Go to the Live Demo
- Click "Try Web Demo"
- Grant microphone permissions
- Start speaking requirements (e.g., "Build a to-do app with dark mode...")
- Watch the PRD build itself in real-time.
For the full deeply integrated experience:
- Download: Get the latest
DEVDRAFT.zipfrom the Landing Page. - Install:
- Open
chrome://extensions - Enable Developer Mode (top right)
- Click Load Unpacked
- Select the unzipped folder
- Open
- Run: Open the Side Panel in any tab to start analyzing.
- Node.js 18+
- Python 3.10+
- Docker (optional)
The backend is cloud-native but can be run locally.
cd backend
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
# Keys required in .env: DEEPGRAM_API_KEY, GEMINI_API_KEY
uvicorn main:app --reloadcd landing
npm install
npm run devcd extension
npm install
npm run build
# Load ./dist into Chrome- Hearing (Deepgram Nova-2): Streams audio to text with <300ms latency.
- Understanding (Gemini 1.5 Pro):
- Analyzes transcripts in 30-word chunks.
- Maintains a "Concept Graph" to detect conflicts.
- Instruction Hierarchy: New instructions with higher timestamps override older conflicting nodes.
- Building (Gemini 2.0 Flash):
- Planning Phase: Architects the folder structure.
- Coding Phase: parallelized file generation.
MIT License. Built for the future of coding.
