SR&ED GPT

Project info

This project is a vendor-agnostic implementation of SR&ED GPT, featuring:

• Frontend: React + Vite (deployed on Vercel)
• Backend: Supabase
• AI: Hybrid 3-Tier Architecture (DeepSeek-R1-Distill-Qwen-1.5B)
  • Tier 1: Hugging Face Serverless API (Primary)
  • Tier 2: Self-Hosted Docker on HF Spaces (Backup)
  • Tier 3: Groq API (Fallback)

Getting Started

Prerequisites

• Node.js & npm
• Supabase CLI
• Docker (optional, for local testing)

Installation

1. Clone the repository:

   git clone https://github.com/TomCruiseTorpedo/rambo2
   cd rambo2

2. Install dependencies:

   npm install

3. Set up environment variables: Create a .env file with:

   VITE_SUPABASE_URL=your_supabase_url
   VITE_SUPABASE_PUBLISHABLE_KEY=your_supabase_key
   

4. Start the development server:

   npm run dev

Deployment

Frontend (Vercel)

Connect this repository to Vercel and deploy. Ensure the build command is npm run build and output directory is dist.

AI (Hugging Face Spaces)

This project uses a Tier 2 Backup hosted on Hugging Face Spaces.

1. Create a new Space (SDK: Docker, Hardware: CPU Basic).
2. Use the helper script to deploy the Dockerfile:

   ./scripts/deploy_to_hf.sh <your_hf_username> <space_name>

   Example: ./scripts/deploy_to_hf.sh myuser tomcruisemissile-rambo2

Backend (Supabase)

Deploy the Edge Functions:

npx supabase functions deploy process-sred --no-verify-jwt
npx supabase functions deploy fill-pdf-t661 --no-verify-jwt

Ensure you have set the following secrets in Supabase:

• LLM_API_URL (Your HF Space URL)
• LLM_API_KEY (Your HF Token)
• GROQ_API_KEY (Optional, for Tier 3 fallback)