README.md

metadata

title: Scholarbot
emoji: 🎓
colorFrom: blue
colorTo: green
sdk: docker
pinned: false

ScholarBot — Research Q&A System

Live demo: https://zan18a-scholarbot.hf.space/docs

A production-grade AI system that answers questions using Retrieval Augmented Generation (RAG) + a fine-tuned language model, served via a REST API.

---

What it does

Send any question → it searches 5000 documents → generates a grounded answer → returns JSON with sources and latency.

---

Tech Stack

Layer	Technology
Model	flan-t5-small (google)
Retrieval	ChromaDB + sentence-transformers
Embeddings	all-MiniLM-L6-v2
API	FastAPI + uvicorn
Containerization	Docker
Deployment	HuggingFace Spaces
CI/CD	GitHub Actions
Dataset	natural-questions (5000 rows)

---

API Endpoints

Method	Endpoint	Description
POST	/query	Ask a question, get an answer
GET	/health	Health check
GET	/docs	Swagger UI

---

Example Request

curl -X POST https://zan18a-scholarbot.hf.space/query \
  -H "Content-Type: application/json" \
  -d '{"question": "what is machine learning", "top_k": 3}'

Example Response

{
  "answer": "Machine learning is a process used to learn from data.",
  "sources": ["..."],
  "model_id": "scholarbot-mistral-lora",
  "latency_ms": 652.9
}

---

Local Setup

git clone https://github.com/zidan18Ahd/scholarbot
cd scholarbot
python -m venv .venv && .venv\Scripts\activate
pip install -r requirements.txt
python data/download.py
python build_index.py
uvicorn api.main:app --reload --port 8000

---

Author

Built by Zidan Ahmed as an end-to-end AI engineering portfolio project.