RAG Chatbot API

A context-only Retrieval-Augmented Generation (RAG) chatbot built with FastAPI. It allows users to upload documents, which are ingested into a hybrid retrieval pipeline (BM25 + dense + cross-encoder reranking + HyDE). When a user asks a question, the system retrieves relevant chunks and uses a local LLM to generate grounded responses based solely on those chunks.

---

Features

• 📁 Multi-format ingestion: Supports .pdf, .docx, .txt, .md
• 📚 Hybrid retrieval: Combines BM25, dense embeddings, HyDE generation, and RRF fusion
• 🧠 Context-only LLM: Answers strictly from provided documents (no hallucination)
• 🧾 Cross-encoder reranking: Improves retrieval quality
• 🔁 Auto reindexing on document upload/delete
• 🖼️ Frontend support: Serves static HTML/JS/CSS

---

📦 Installation

1. Clone the Repository

git clone https://github.com/your-username/rag-chatbot-api.git
cd rag-chatbot-api

2. Create a virtual environment

python -m venv venv
source venv/bin/activate  # or venv\Scripts\activate on Windows

if prefer using Conda

conda create --name <env_name> python=<version>
conda activate <env_name>

3. Installing dependencies

pip install -r requirements.txt

Project Structure

.
├── app.py # FastAPI app entrypoint
├── bulk_ingest.py # Ingests all files from ./data
├── data/ # Uploaded documents
├── static/ # Frontend HTML/CSS/JS
├── models/
│ └── llama_wrapper.py # Local LLM (Mistral/OpenHermes wrapper)
├── memory/
│ ├── embedder.py # Embedding logic (MiniLM or similar)
│ ├── memory_store.py # Vector store (e.g., ChromaDB)
│ └── logger.py # JSON-based chat logger
├── utils/
│ ├── pipeline_runner.py # HyDE and standard pipeline runners
│ └── prompt_builder.py # Builds system+context prompts
├── hybrid_retrieval_pipeline.py# Hybrid RAG retrieval logic
├── document_ingestor.py # Ingests individual documents
├── requirements.txt
└── README.md

Running the App

uvicorn app:app --reload

API endpoints

Method	Endpoint	Description
GET	/	Serves the frontend UI
POST	/upload	Uploads and ingests a document
POST	/chat	Chat with the LLM using RAG
GET	/documents	Lists all uploaded documents
DELETE	/documents/{fn}	Deletes a document and reindexes
POST	/reindex	Manually reindex all files
GET	/health	Health check for readiness

LLM + Retrieval

LLM: openhermes-2.5-mistral-7b.Q4_K_M.gguf (run locally via llama-cpp-python)

Embedding: all-MiniLM-L6-v2 via SentenceTransformers

Cross-Encoder: ms-marco-MiniLM-L-6-v2

Vector store: ChromaDB (can be swapped)

Sample screenshots