🧠 AskDoc Local - RAG API with FastAPI + Ollama + LangChain

AskDoc Local is a fully offline, API-first Retrieval-Augmented Generation (RAG) system built using:

• 🧩 FastAPI – For REST APIs
• 💬 Ollama – For running lightweight LLMs like mistral locally
• 🔗 LangChain – For document chunking, embedding, and querying
• 📄 PDF/CSV/XLSX/DOCX – Multi-format document support
• 🧠 Chroma (or replaceable) – Local vector store (swappable with FAISS or Weaviate)

---

📦 Features

Feature	Status
Upload documents (PDF, CSV, XLSX, DOCX)	✅
Parse and chunk text using LangChain	✅
Embed using Ollama models (e.g., mistral)	✅
Store embeddings locally in ChromaDB	✅
Ask questions via /query/ API	✅
Delete single or all documents	✅
Debug API to inspect indexed documents	✅
Reset API to wipe everything clean	✅

---

🚀 Quick Start

1. Clone & Install

git clone https://github.com/sonwanesuresh95/askdoc-local.git
cd askdoc-local
pip install -r requirements.txt

2. Start Ollama

ollama run mistral

Make sure Ollama is running and the model is pulled.

3. Start the API server

uvicorn app.main:app --reload

---

🛠️ API Endpoints

📄 Upload Document

POST /upload/
Content-Type: multipart/form-data

Upload .pdf, .docx, .csv, or .xlsx. The document is parsed, chunked, embedded, and stored.

---

❓ Ask a Question

POST /query/
{
  "query": "What is Acme Corp's revenue?"
}

Returns an answer based on uploaded and embedded documents.

---

🗑️ Delete a Document

DELETE /api/documents/{doc_id}

Deletes uploaded file, parsed .txt, and embeddings.

---

🔄 Reset Entire System

POST /reset/

Deletes all uploaded files, parsed text, and vector DB.

---

🐞 Debug Vector Store

GET /debug/vectorstore

Returns a list of indexed documents and their chunk counts.

---

📁 Directory Structure

askdoc-local/
├── app/
│   ├── main.py
│   ├── routes/
│   ├── services/
├── data/
│   ├── documents/         # Uploaded files
│   ├── parsed_text/       # Parsed .txt files
│   ├── vector_store/      # Chroma DB data
│   └── document_index.json

---

🧠 Embedding Model

• Uses OllamaEmbeddings(model="mistral") from langchain_ollama
• You can replace with any local model supported by Ollama (e.g., llama2, codellama, phi)

---

🔄 Future Plans

• Frontend UI using React + Material UI
• Weaviate-based vector backend
• Docker + CI/CD setup
• Streamlit Playground UI

---

🧑‍💻 Author

Suresh Sonwane
GitHub: @sonwanesuresh95

---

📜 License

MIT License