Yantr — AI Lab Assistant

Yantr is an AI-powered study assistant built for engineering students. It combines a RAG (Retrieval Augmented Generation) pipeline with an organized library of study materials, enabling precise, context-aware answers sourced directly from indexed course documents.

---

Features

Feature	Description
RAG Chatbot	Streams GPT-4o-mini responses grounded in your indexed PDFs via Pinecone
Study Library	Organized browser for PDFs — Physics, Chemistry, C Programming, CPS, Statistics, FCPS
Vision Support	Upload images alongside questions for multimodal AI assistance
Contribution Pipeline	Students submit PDFs via UI → GitHub PR auto-created → Admin merges to publish
Auth & Quotas	Clerk authentication + Upstash Redis rate limiting (10 messages/day)
Streaming Responses	Real-time streamed AI output using Vercel AI SDK

---

Tech Stack

Layer	Technology
Frontend	Next.js 16, React 19, TypeScript
Styling	Tailwind CSS v4, Framer Motion
AI / LLM	OpenAI GPT-4o-mini (@ai-sdk/openai)
Vector DB	Pinecone (text-embedding-3-small)
Auth	Clerk
Rate Limiting	Upstash Redis
Contribution API	Octokit (@octokit/rest) → GitHub REST API
Deployment	Vercel
Python Backend	pdfplumber, langchain, openai (for RAG indexing)

---

Repository Structure

Yantr-chatbot/
├── frontend/                    # Next.js application
│   ├── src/
│   │   ├── app/
│   │   │   ├── api/
│   │   │   │   ├── chat/        # Streaming RAG chat endpoint
│   │   │   │   ├── contribute/  # GitHub PR creation endpoint
│   │   │   │   ├── history/     # Chat history (Upstash Redis)
│   │   │   │   ├── quota/       # Daily message quota API
│   │   │   │   └── theory/      # Study material file server
│   │   │   ├── contribute/      # User contribution form page
│   │   │   ├── resources/       # Study materials browser page
│   │   │   └── page.tsx         # Home / Chat interface
│   │   ├── components/          # Reusable React components
│   │   └── lib/                 # Pinecone, Upstash, Clerk helpers
│   ├── public/
│   │   └── manuals/             # PDFs served to users
│   └── package.json
│
└── backend/                     # Python data processing
    ├── scripts/
    │   ├── upload_vectors.py    # Indexes PDFs into Pinecone
    │   └── generate-theory-index.js
    └── data/
        └── knowledge_base/      # Source documents (FCPS, Stats, CPS...)

---

Getting Started

Prerequisites

• Node.js 18+
• Python 3.9+ (for indexing only)

1. Clone & Install

git clone https://github.com/aryawadhwa/Yantr-chatbot.git
cd Yantr-chatbot/frontend
npm install

2. Environment Variables

Create frontend/.env.local:

# AI
OPENAI_API_KEY=sk-...

# Vector DB
PINECONE_API_KEY=...
PINECONE_INDEX_NAME=yantr-knowledge-base

# Auth
NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=pk_...
CLERK_SECRET_KEY=sk_...

# Rate Limiting
UPSTASH_REDIS_REST_URL=https://...
UPSTASH_REDIS_REST_TOKEN=...

# Contribution Pipeline (GitHub Fine-grained PAT)
# Scopes required: Contents (read/write), Pull Requests (read/write)
GITHUB_PAT=github_pat_...

3. Run Development Server

cd frontend
npm run dev

Open http://localhost:3000.

---

RAG Indexing (One-Time Setup)

To embed course documents into Pinecone:

# Setup Python environment
cd backend
python -m venv venv
source venv/bin/activate  # macOS/Linux
pip install pdfplumber langchain-text-splitters openai pinecone python-dotenv

# Add PDFs to backend/data/knowledge_base/
# Then run the indexer
python scripts/upload_vectors.py

---

Contribution Pipeline

Students can submit study materials directly from the web UI:

1. Navigate to /contribute on the site.
2. Select a course, enter a filename, and upload a PDF.
3. The server calls the GitHub API to create a new branch and PR with the file placed in both public/manuals/ and data/knowledge_base/.
4. An admin reviews and merges the PR on GitHub — Vercel auto-deploys the update.

---

Deployment (Vercel)

1. Push code to GitHub.
2. Import the repository on vercel.com.
3. Set Root Directory to frontend.
4. Add all environment variables from .env.local in Project Settings.
5. Click Deploy.

---

Team

Name	Role
Arya Wadhwa	CEO & Co-Founder
Dilraj Singh	CTO & Co-Founder
Aadarsh KR	Co-Founder
Harsha S	Co-Founder