AstroGPT 🚀

AstroGPT is a learning-oriented project built with FastAPI, FAISS, SentenceTransformers, and Ollama.
The main goal is to practice the fundamentals of Retrieval-Augmented Generation (RAG), experiment with FAISS for vector search, and integrate a local LLM through Ollama. Data source : https://www.kaggle.com/datasets/patrickfleith/astrochat

---

🛰 Technologies Used

• FastAPI → Backend REST service
• FAISS → Vector search engine
• SentenceTransformers → Embedding generation for queries and documents
• Ollama (LLaMA 3.1:8B) → Language model
• HTML + Vanilla JS → Frontend interface
• pandas / numpy → Data processing

---

⚙️ System Architecture

1. Frontend (chat.html) → Collects user messages, sends them to the backend, and renders responses.
2. Backend (FastAPI)
   • Generates embeddings for user queries.
   • Retrieves the most relevant chunks using FAISS.
   • Builds an augmented prompt and sends it to Ollama.
   • Returns the model’s response to the frontend.
3. Ollama → Generates the final answer using the retrieved context.

---

🚀 How to Run

1. Install dependencies:

   pip install -r requirements.txt
   

2. Start Ollama and pull the model:

   ollama run llama3.1:8b
   

3. Run the FastAPI app:

   uvicorn app:app --reload --port 8000
   

4. Open in your browser:

   http://localhost:8000