🐼 PandaPrep — Agentic AI Revision & Study Notes Platform

Live Application: https://pandaprep.vercel.app/

PandaPrep is an agentic study notes generation and interactive learning platform built for university students and educators. It transforms complex, multi-unit syllabi and reference materials into mathematically rigorous, beautifully structured revision guides with LaTeX formulas, diagrams, and self-correcting verification.

---

🏗️ Monorepo Architecture

PandaPrep is organized as a lightweight monorepo containing three core services:

pandaprep-live/
├── frontend/             # Next.js 15 (App Router, TailwindCSS, KaTeX, Client-side PDF export)
├── backend-agentic/      # Bounded LangGraph.js Agentic Runtime in TypeScript
├── backend/              # User account management, Razorpay billing, Cloudinary storage
├── vercel.json           # Multi-service monorepo routing configuration
├── full-architecture.md  # Detailed system architecture blueprint
└── agentic-observability-evals-report.md # Observability, metrics, and evaluation engineering report

Service Breakdown

Service	Technology	Role & Key Responsibilities
frontend	Next.js 15, React 19, TailwindCSS, KaTeX	Responsive student portal, interactive Markdown/LaTeX viewer, real-time generation polling, client-side vector PDF generation (window.print()).
backend-agentic	TypeScript, LangGraph.js, Express, MongoDB Atlas	Bounded stateful agent pipeline: Intake $\to$ Planner DAG $\to$ Scoped Writers $\to$ Verifier (6 contract checks) $\to$ Targeted Repair Loop $\to$ Finalize.
backend	Node.js, Express, Mongoose, Razorpay SDK	User authentication checks, subscription credits, payment webhooks, and PDF upload to Cloudinary.

---

🤖 Agentic Generation Workflow (backend-agentic)

Unlike traditional linear automation scripts, PandaPrep executes a bounded stateful agent graph powered by LangGraph.js and MongoDB Atlas checkpoints:

Resilient Multi-Provider LLM Tiering

1. Primary: Google Gemini (gemini-3.5-flash-lite)
2. Fallback 1: Groq (openai/gpt-oss-20b)
3. Fallback 2: OpenCode Zen (deepseek-v4-flash-free)
4. Vector Embeddings: Google Gemini (gemini-embedding-2, 768-dim normalized)

---

🚀 Getting Started (Local Development)

Prerequisites

• Node.js v20+
• MongoDB Atlas Cluster (Free Tier M0 or higher)
• API Keys for Gemini, Firebase, and optional observability (Langfuse)

1. Setup Environment Variables

• Frontend: Create frontend/.env.local

  NEXT_PUBLIC_FIREBASE_API_KEY="AIza..."
  NEXT_PUBLIC_FIREBASE_AUTH_DOMAIN="pandaprep-22edc.firebaseapp.com"
  NEXT_PUBLIC_FIREBASE_PROJECT_ID="pandaprep-22edc"
  NEXT_PUBLIC_RAZORPAY_KEY_ID="rzp_test_..."
  NEXT_PUBLIC_PROD_BASE_URL="http://localhost:8000/"
  NEXT_PUBLIC_AGENTIC_BASE_URL="http://localhost:8001/api"

• Agentic Backend: Create backend-agentic/.env

  PORT=8001
  NODE_ENV=development
  MONGODB_URI="mongodb+srv://<user>:<pass>@cluster.mongodb.net/pandaprep"
  GEMINI_API_KEY="AIza..."
  OPENCODE_API_KEY="opencode_..."
  LANGFUSE_PUBLIC_KEY="pk-lf-..."
  LANGFUSE_SECRET_KEY="sk-lf-..."
  LANGFUSE_HOST="https://cloud.langfuse.com"

• Legacy Backend: Create backend/.env

  PORT=8000
  MONGODB_URI="mongodb+srv://<user>:<pass>@cluster.mongodb.net/pandaprep"
  CLOUDINARY_CLOUD_NAME="..."
  CLOUDINARY_API_KEY="..."
  CLOUDINARY_API_SECRET="..."
  RAZORPAY_KEY_ID="rzp_test_..."
  RAZORPAY_KEY_SECRET="..."

2. Run All Services

In three separate terminal windows:

# Terminal 1: Frontend
cd frontend
npm install
npm run dev

# Terminal 2: Agentic Backend
cd backend-agentic
npm install
npm run dev

# Terminal 3: Legacy Backend
cd backend
npm install
npm run dev

• Frontend runs on http://localhost:3000
• Agentic Backend runs on http://localhost:8001
• Legacy Backend runs on http://localhost:8000

---

🧪 Testing & Automated Evals

Run the test harness inside backend-agentic:

# Run unit, contract, and integration tests (Vitest)
npm test

# Run LLM-as-a-Judge offline golden dataset evaluation benchmark
npm run test:evals

---

🚢 Deployment on Vercel

PandaPrep is configured for zero-friction Vercel deployment:

1. Multi-Service Monorepo: Connect your GitHub repository to Vercel and select the Services preset. The root vercel.json routes /api/backend to backend, /api/agentic to backend-agentic, and all other traffic to frontend.
2. Independent Deployments: Alternatively, import frontend, backend, and backend-agentic as three separate Vercel projects setting their respective Root Directories.

---

📄 License

ISC License. Built for students with ❤️ by Team PandaPrep.