LeetCode Assistant

Productivity assistant for LeetCode that lives directly inside the browser. A Chrome extension observes the active problem, syncs structured context to a FastAPI backend, runs retrieval-augmented generation (RAG) with vector search + LLMs, and returns curated guidance in a side panel chat.

---

✨ Features

• In-browser coaching – context-aware chat panel that understands the LeetCode problem you currently have open.
• Automatic problem scraping – DOM + GraphQL scrapers populate description, constraints, and examples with millisecond latency.
• RAG pipeline – LangGraph orchestrates pgvector similarity search, Redis caching, and an LLM prompt that returns JSON summaries.
• Streaming answers + syntax highlighting – responses arrive token-by-token and render rich Markdown with Prism-powered code blocks.
• Theme toggle built in – switch between light and dark modes on the fly; the extension remembers your preference.
• Modular monorepo – separate apps/extension and apps/api packages with shared types and reproducible builds.
• Developer tooling – Docker Compose stack, hot-reload Uvicorn server, npm proxy scripts, and detailed setup docs.

---

🧱 Architecture

├── apps
│   ├── api
│   │   ├── app
│   │   │   ├── api         # FastAPI routers
│   │   │   ├── core        # Settings & logging
│   │   │   ├── db          # SQLAlchemy engine + pgvector models
│   │   │   ├── schemas     # Pydantic request/response models
│   │   │   └── services    # RAG pipeline, caching, embeddings
│   │   ├── Dockerfile
│   │   ├── requirements.txt
│   │   └── .env.example
│   └── extension
│       ├── public
│       ├── src             # React/Tailwind Chrome extension source
│       ├── dist            # Built assets (generated)
│       ├── package.json
│       └── vite.config.ts
├── infra
│   └── docker
│       └── docker-compose.yml
└── package.json            # Root scripts (proxy to extension workspace)

---

🔑 Tech Stack

• Frontend: React 18, TypeScript, Vite, Tailwind CSS, Chrome Extension APIs
• Backend: FastAPI, AsyncIO, Pydantic, LangChain, LangGraph, OpenAI API
• Data Layer: PostgreSQL 16 (pgvector), Redis 7, SQLAlchemy 2.x
• Infrastructure: Docker, Docker Compose, uvicorn/gunicorn

---

🚀 Quick Start

1. Prerequisites

• Node.js 18+
• Docker Desktop (or Docker Engine) + Docker Compose V2
• An OpenAI API key (for both chat + embeddings)

2. Clone and install

git clone https://github.com/<you>/leetcode-assistant.git
cd leetcode-assistant

# Install extension dependencies
npm install --prefix apps/extension

3. Configure the backend

cp apps/api/.env.example apps/api/.env
# edit apps/api/.env and set:
# OPENAI_API_KEY=sk-your-secret-key

4. Run backend services

docker compose -f infra/docker/docker-compose.yml up -d --build
curl http://localhost:8000/api/v1/health   # → {"status":"healthy"}

Services exposed locally:

Service	Port	Notes
FastAPI	8000	REST API & docs at /docs
Postgres	5432	Includes pgvector
Redis	6379	Used for response caching

5. Build and load the Chrome extension

npm run build --prefix apps/extension

• Open Chrome → chrome://extensions
• Enable Developer mode
• Click Load unpacked and choose apps/extension/dist/
• Open the extension’s Settings page, confirm the API base URL (http://localhost:8000/api/v1), click Save and Test connection
• Navigate to any LeetCode problem and open the LeetCode Assistant side panel

---

🛠️ Development Scripts

From repository root:

Command	Action
npm run dev	Start the Vite dev server for the extension (apps/extension)
npm run build	Type-check + production build of the extension
npm run lint	ESLint for the extension codebase
python3 -m compileall apps/api/app	Quick syntax check for backend modules

Backend container uses live-reload via Uvicorn when /app/app changes.

Tests

• Backend: pytest apps/api/tests
• Extension: build + lint (no unit tests yet)

Continuous Integration

GitHub Actions (.github/workflows/ci.yml) runs on every push/PR to main. The workflow builds the FastAPI service (pip install + python -m compileall) and verifies the Chrome extension (npm ci, lint, and build). Keep new dependencies declared in apps/api/requirements.txt or apps/extension/package.json so the pipeline succeeds.

---

🧠 RAG Pipeline Overview

1. Problem ingestion – extension posts the scraped problem to /api/v1/documents.
2. Embedding + storage – FastAPI service chunkifies content, obtains embeddings, and writes vectors to PostgreSQL (pgvector).
3. Retrieval – when a chat question arrives, LangGraph queries pgvector for top-k chunks, enriched with metadata.
4. LLM response – LangChain ChatOpenAI consumes a structured prompt and returns JSON containing answer + summary.
5. Caching – Redis stores chat responses keyed by slug + normalized conversation history to avoid repeated LLM calls.

---

🧪 API Reference

• GET /api/v1/health – readiness probe
• GET /api/v1/docs – Swagger UI
• POST /api/v1/documents – ingest problem context (handled by extension)
• POST /api/v1/chat – accepts question, problem, and history, returns answer + summary + source snippets

---

🛡️ Troubleshooting

Symptom	Fix
Extension shows “Backend unreachable”	Ensure Docker stack is running and API URL is http://localhost:8000/api/v1; click Test connection in Settings.
“Context sync failed”	The /documents endpoint rejected the request. Check logs (docker compose … logs api) for authentication or schema errors.
“Failed to fetch” after sending a chat	Inspect API logs for 500 errors (often due to invalid JSON shape or missing API key).
Permission denied to Docker socket	On macOS, grant Terminal/CLI access to Docker Desktop or rerun with elevated privileges.

---

🗺️ Roadmap Ideas

• Streaming responses for richer UX
• Persistent user chat history
• Multi-user authentication / rate limiting
• LangSmith or OpenTelemetry tracing for prompt evaluation
• Gemini or other provider fallbacks

---

With the backend running and the extension loaded, open any LeetCode problem and launch the side panel—LeetCode Assistant will ingest the context, surface relevant snippets, and deliver actionable guidance tailored to your question. Happy grinding! 🙌