📚 TDS Virtual Teaching Assistant

This project implements a FastAPI-based virtual teaching assistant that answers questions using IITM BSc course material (e.g., course notes, discourse discussions). It supports both text queries and returns relevant answers and source links.

---

🚀 Features

• Accepts a question via a REST API (/api/)
• Retrieves the most relevant answers from course content
• Returns a concise answer and list of relevant links
• Hosted using a static ngrok domain:
  🌐 https://eel-saving-barely.ngrok-free.app/api/

---

📦 Endpoint

POST /api/

Request JSON Body

{
  "question": "How does bagging improve decision tree performance?"
}
{
  "answer": "Bagging improves decision tree performance by reducing variance through ensembling...",
  "links": [
    {
      "url": "https://discourse.onlinedegree.iitm.ac.in/t/bagging-vs-boosting",
      "text": "Discussion comparing bagging and boosting"
    },
    {
      "url": "https://notes.iitm.ac.in/ml/week6",
      "text": "Notes: Week 6 - Ensemble Methods"
    }
  ]
}

🛠 How It Works

 - Loads course content from data/all_documents.json
 - Embeds them using OpenAI embeddings
 - Answers user questions using semantic similarity search
 - Returns the top match with supporting links

📂 Repo Structure

tds-virtual-ta/
├── data/
│   └── all_documents.json       # All course content (notes + discourse)
├── embed_and_search.py         # Embedding + search logic
├── main.py                     # FastAPI server and endpoint logic
├── requirements.txt            # Python dependencies
└── README.md                   # This file

✅ To Run (Colab Friendly)

1. Install required packages:

!pip install fastapi uvicorn nest_asyncio pyngrok

2. Clone this repo and start the server using ngrok:

   Just run the following code in colab : thank me later :)

!pip install fastapi uvicorn nest_asyncio pyngrok --quiet

# Remove existing folder if exists, then clone
import shutil
import os

if os.path.exists("tds-virtual-ta"):
  shutil.rmtree("tds-virtual-ta")

import os
os.environ["AI_PROXY_API_KEY"] = "replace with a token from aiproxy.sanand.workers.dev"


!git clone --depth=1 https://github.com/sanaIITM/tds-virtual-ta.git
%cd tds-virtual-ta

from pyngrok import ngrok
import nest_asyncio
import uvicorn
import main  # assumes main.py is in tds-virtual-ta

# Step 4: Connect with static domain from ngrok
ngrok.set_auth_token("replace with you ngrok auth token") #get your auth token from https://dashboard.ngrok.com/get-started/your-authtoken
public_url = ngrok.connect(addr=8000, domain="replace with your subdomain ") # under domain (https://dashboard.ngrok.com/domains) select create domain and get your url
print("🌐 Public URL:", public_url)

# Step 5: Run app
nest_asyncio.apply()
uvicorn.run(main.app, host="0.0.0.0", port=8000)

3. Access Swagger UI at: [https://eel-saving-barely.ngrok-free.app/docs]

This API is meant for the TDS Project - Virtual TA (May 2025).

👤 Author

Sana Bint Salim