Study Desk API

A FastAPI backend for the Study Desk app. Upload notes or a textbook chapter, and it uses OpenAI for embeddings + generation and Pinecone as the vector store, so each generation step (concepts, questions, flashcards, study plan) pulls the most relevant passages via retrieval (RAG) instead of always stuffing the whole document into the prompt.

How it works

1. POST /api/upload (or /api/upload-text for pasted text) — extracts text from a .txt/.pdf file, splits it into ~700-token chunks, embeds each chunk with text-embedding-3-small, and stores them in a Pinecone namespace unique to that upload (session_id).
2. POST /api/generate/{kind} where kind is one of concepts, questions, flashcards, plan — embeds a task-specific retrieval query (e.g. "key terms and definitions"), fetches the top matching chunks for that session_id from Pinecone, and asks an OpenAI chat model to generate structured JSON from just that context.
3. DELETE /api/session/{session_id} — deletes that session's vectors from Pinecone once you're done with it.

Setup

python3 -m venv .venv
source .venv/bin/activate          # Windows: .venv\Scripts\activate
pip install -r requirements.txt

cp .env.example .env
# then edit .env and add your OPENAI_API_KEY and PINECONE_API_KEY

Run it:

uvicorn app.main:app --reload

The API is now at http://localhost:8000, with interactive docs at http://localhost:8000/docs.

The Pinecone index is created automatically on first run (serverless, cosine similarity, matching the embedding model's dimension) if it doesn't already exist.

Example usage

# Upload a file
curl -X POST http://localhost:8000/api/upload \
  -F "file=@chapter3.pdf"
# -> {"session_id": "...", "chunk_count": 12, ...}

# Or upload pasted text
curl -X POST http://localhost:8000/api/upload-text \
  -H "Content-Type: application/json" \
  -d '{"text": "Photosynthesis is the process by which..."}'

# Generate flashcards for that session
curl -X POST http://localhost:8000/api/generate/flashcards \
  -H "Content-Type: application/json" \
  -d '{"session_id": "PASTE_SESSION_ID_HERE"}'

# kind can also be: concepts | questions | plan

# Clean up when done
curl -X DELETE http://localhost:8000/api/session/PASTE_SESSION_ID_HERE

Project layout

app/
  main.py                 FastAPI app, CORS, router registration
  config.py                Settings loaded from environment / .env
  models.py                 Pydantic request/response schemas
  services/
    extraction.py            Pulls text out of .txt / .pdf uploads
    chunking.py               Token-aware chunking (tiktoken)
    embeddings.py             OpenAI embeddings, batched
    vectorstore.py            Pinecone index create / upsert / query / delete
    generation.py             Retrieval + OpenAI chat generation per task
  routers/
    upload.py                 /api/upload, /api/upload-text
    generate.py                /api/generate/{kind}, /api/session/{id}
requirements.txt
.env.example

Notes for production use

This is a working starting point, not a hardened production service. Before shipping it for real users, you'll likely want to add:

• Auth on the endpoints (API key, JWT, or your app's session auth).
• Rate limiting, since each generation call costs OpenAI tokens.
• Persistent session metadata (right now the only record of a session is its Pinecone namespace — consider a small database table mapping session_id → user, filename, upload time, so you can list/expire old uploads).
• A scheduled cleanup job to delete old Pinecone namespaces that were never explicitly deleted.
• Tightening ALLOWED_ORIGINS in .env to your actual frontend's origin instead of *.

Connecting a frontend

Any frontend can talk to this by calling /api/upload (or /api/upload-text) to get a session_id, then calling /api/generate/{kind} for each of concepts, questions, flashcards, and plan with that session_id. If you're using the Study Desk HTML artifact built earlier, swap its direct

DISCLAIMER: THIS APP IS UNDER TESTING PHASE ,SO IF ANYONE -USING THIS CODE, FINDS ANY ERROR THEN PLEASE DO SEND ME A PULL REQUEST MENTIONING THE SAME.

fetch() calls to api.anthropic.com for calls to this backend's endpoints instead.