https://github.com/dark-ang3l07/nucleus-rag

A production-grade, multi-model RAG (Retrieval-Augmented Generation) system built for resilience and accuracy, featuring a TypeScript/React frontend and a Python/FastAPI backend. Architected to run efficiently on Hugging Face Spaces.
https://github.com/dark-ang3l07/nucleus-rag

docker-compose gemini-api google huggingface langchain

Last synced: 3 months ago
JSON representation

A production-grade, multi-model RAG (Retrieval-Augmented Generation) system built for resilience and accuracy, featuring a TypeScript/React frontend and a Python/FastAPI backend. Architected to run efficiently on Hugging Face Spaces.

• Host: GitHub
• URL: https://github.com/dark-ang3l07/nucleus-rag
• Owner: daRk-ang3L07
• Created: 2026-04-02T08:30:05.000Z (3 months ago)
• Default Branch: main
• Last Pushed: 2026-04-02T15:16:52.000Z (3 months ago)
• Last Synced: 2026-04-03T03:40:02.239Z (3 months ago)
• Topics: docker-compose, gemini-api, google, huggingface, langchain
• Language: TypeScript
• Homepage: https://dark-ang3l-nucleus-rag.hf.space/#
• Size: 1.05 MB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          

# 🚀 Nucleus RAG: High-Performance GenAI Pipeline

A production-grade, multi-model RAG (Retrieval-Augmented Generation) system built for resilience and accuracy, featuring a **TypeScript/React** frontend and a **Python/FastAPI** backend. Architected to run efficiently on **Hugging Face Spaces**.

## 🏆 Industrial-Strength Features

- **Multi-Model Fallback Router**: Intelligent failover from **Google Gemini 2.5** to **Hugging Face (Qwen-7B)** during API quota (429) events, ensuring 100% chat availability.

- **Advanced Retrieval Pipeline**: Orchestrated via **LangChain**, combining **Hybrid Search (Keyword + Semantic)** with **Multi-Query Expansion** and **TinyBERT Cross-Encoder Reranking**.

- **Quota-Friendly "Lite Auditor"**: A custom-engineered evaluation suite that performs full RAG benchmarking (Faithfulness, Relevancy) in a single-shot AI call to survive limited API quotas.

- **TypeScript Glassmorphism UI**: A premium dashboard featuring citation accordions, real-time status polling, and markdown-rendered AI responses.

- **Dockerized Architecture**: Fully containerized for one-click deployment to any OCI-compliant cloud provider.

---

## 🛠️ Tech Stack

- **AI/LLM**: Google Gemini 2.5 Flash, LangChain, Qwen-7B (Hugging Face), Phi-3 Mini.

- **Database**: ChromaDB (Vector Store), Supabase (Chat History & Auth).

- **Backend**: FastAPI, Asynchronous Python, Background Tasks, Docker.

- **Frontend**: React 18, TypeScript, Vite, Framer Motion, TailwindCSS.

---

## 🚀 One-Click HF Spaces Setup

To deploy this project to **Hugging Face Spaces**:

1. **Create a New Space**: Select the **Docker** SDK.

2. **Configure Secrets**: Add the following variables to your Space settings:

   - `GOOGLE_API_KEY`: Your Gemini API Key.

   - `HUGGINGFACEHUB_API_TOKEN`: For reranking and model fallbacks.

   - `SUPABASE_URL` & `SUPABASE_ANON_KEY`: For chat persistence.

3. **Push to Main**: The `Dockerfile` is pre-configured to handle the multi-process environment.

---

## 🧪 AI Performance Monitoring

The system includes a dedicated `/api/v1/evaluate/` endpoint that benchmarks your RAG pipeline. It uses a specialized "Lite Auditor" prompt to survive the 20-request daily limit of experimental LLM tiers.

**Key Metrics:**

- **Faithfulness**: Mathematically proves the answer is derived ONLY from your documents.

- **Answer Relevancy**: Measures how precisely the AI addressed the user's specific intent.