Choti Portfolio API — RAG Chatbot (Gemini)

RAG-powered serverless API that answers questions about me and my work. It uses a lightweight in-memory vector store loaded from precomputed embeddings and Google Gemini for both embeddings and generation.

---

The Story Behind This API

Because "Please Please Please" Don't Give Me Generic Responses

Just like Sabrina sings about wanting something real — this API delivers context-rich answers about my portfolio instead of the same old chatbot nonsense. 🎵

✨ Features

• 🤖 RAG answers: Retrieves relevant snippets from contact details/profile/project documents
• 🧠 Gemini models: text-embedding-004 for search, configurable chat model for generation
• 🚀 Simple API: POST /api/chat-rag with optional conversation history
• ☁️ Serverless ready: Vercel functions with permissive CORS headers
• 🔍 Vector similarity search: Cosine similarity-based document retrieval
• 🛡️ Health monitoring: Built-in health check endpoints

🛠️ Tech Stack

• Runtime: Node.js 18+ (ES Modules)
• Framework: Express (for local dev), Vercel Functions (deployment)
• AI/ML: @google/generative-ai
• Vector Store: Custom in-memory implementation with Gemini embeddings
• Deployment: Vercel
• Database: File-based embeddings storage

📋 Prerequisites

Before running this project, make sure you have:

• Node.js 18+ (recommended)
• A Google Gemini API key
• Create .env.local in the project root with:

  GEMINI_API_KEY=your_api_key_here
  # optional, defaults to gemini-2.0-flash-lite
  GEMINI_MODEL=gemini-2.0-flash-lite

🚀 Installation

# Clone the repository
git clone https://github.com/jgchoti/jgchoti-api.git
cd jgchoti-api

# Install dependencies
npm install

💻 Running Locally

Two options available:

npm run dev
# -> http://localhost:3000

The dev server logs available endpoints on start.

📚 API Documentation

Base URL

http://localhost:3000/api

Endpoints

GET /api/health

Health check and basic info.

curl -s http://localhost:3000/api/health | jq

Response:

{
  "status": "healthy",
  "message": "Choti's Portfolio API is running!",
  "endpoints": {
    "chat": "/api/chat-rag",
    "health": "/api/health"
  },
  "timestamp": "2025-09-08T17:44:01.845Z"
}

GET /api/

Root API info and usage documentation.

curl -s http://localhost:3000/api/

Response:

{
  "message": "Welcome to Choti's Portfolio API! 🤖",
  "description": "RAG-powered chatbot API for portfolio inquiries",
  "endpoints": {
    "chat": "/api/chat-rag - POST - Chat with Choti's AI agent",
    "health": "/api/health - GET - Health check"
  },
  "usage": {
    "method": "POST",
    "url": "/api/chat-rag",
    "body": {
      "message": "Your question here",
      "conversationHistory": "Optional previous messages"
    }
  },
  "portfolio": "https://jgchoti.github.io",
  "github": "https://github.com/jgchoti"
}

POST /api/chat-rag

Chat with the RAG agent about Choti. Returns model text and debug metadata.

Example Request:

curl -s -X POST https://jgchoti-api.vercel.app/api/chat-rag \
  -H 'Content-Type: application/json' \
  -d '{
    "message": "Summarize Choti'\''s international experience.",
    "conversationHistory": []
  }'

Response Structure:

{
  "response": "Choti has a truly global perspective, having lived and worked in nine countries, including Thailand, Switzerland, and the UK. This international experience gives her a unique edge in cross-cultural collaboration and understanding global markets. You can see how she applies this in her data projects here: https://jgchoti.github.io/data. What kind of role are you looking to fill?",
  "metadata": {
    "model": "gemini-2.0-flash-lite",
    "ragEnabled": true,
    "vectorUsed": true,
    "vectorDebugInfo": {
      "resultsUsed": 3,
      "topSimilarity": 0.6907635643942781,
      "types": ["profile", "contact", "profile"],
      "similarities": [
        0.6907635643942781, 0.6208553455982836, 0.5978628425932092
      ]
    },
    "contextLength": 3803,
    "timestamp": "2025-09-08T17:48:34.895Z"
  }
}

Notes:

• conversationHistory is optional; items should be { type: 'user' | 'assistant', content: string }
• If embeddings are missing or the vector store is not ready, the endpoint falls back to a small default context

🔍 Embeddings and RAG

The vector store (lib/HybridVectorStore.js) loads from data/embeddings-gemini.json and uses cosine similarity to retrieve relevant snippets. Query embeddings are computed on-the-fly using Gemini text-embedding-004, so GEMINI_API_KEY is required both to generate embeddings and to serve queries.

🔍 Dynamic Portfolio Integration

The system automatically enriches portfolio data through:

• GitHub API integration (scripts/generate-github.js) that discovers and analyzes repositories and content extraction from README files and repository metadata
• Career path relevance scoring for data engineering, data science, ML engineering, and backend development roles

Generate/Refresh Embeddings

npm run generate-embeddings

This script builds content from:

• data/profileData.js
• data/projectData.js
• data/contactInfo.js
• data/github_portfolio_data.json

Creates embeddings with Gemini and writes:

• data/embeddings-gemini.json
• data/embeddings-gemini-metadata.json

🔧 Environment Variables

Create a .env.local file in the project root:

# Required: Google Gemini API key
GEMINI_API_KEY=your_api_key_here

# Optional: Gemini model for generation (defaults to gemini-2.0-flash-lite)
GEMINI_MODEL=gemini-2.0-flash-lite

For Vercel deployment: Set these in the project's Environment Variables dashboard.

📁 Project Structure

jgchoti-api/
├── api/                      # Serverless endpoints for Vercel
│   ├── chat-rag.js          # Main RAG chat endpoint (POST)
│   ├── health.js            # Health check (GET)
│   └── index.js             # Root API info (GET)
├── data/                     # Data and generated embeddings
│   ├── profileData.js       # Profile information
│   ├── projectData.js       # Project data
│   ├── contactInfo.js       # Contact information
│   ├── embeddings-gemini.json          # Generated embeddings
│   └── embeddings-gemini-metadata.json # Embedding metadata
├── lib/
│   └── HybridVectorStore.js # In-memory vector store using Gemini embeddings
├── scripts/
│   └── generate-embeddings.js # Builds embeddings from profile/projects
├── dev-server.js             # Local Express server that proxies to handlers
├── vercel.json               # Vercel config (headers, rewrites, function settings)
├── package.json              # Scripts and dependencies
├── .env.local                # Environment variables (create this)
└── README.md

🚀 Deployment (Vercel)

Step-by-step deployment:

1. Login and link project:

   vercel login
   vercel link

2. Set environment variables:

   vercel env add GEMINI_API_KEY
   vercel env add GEMINI_MODEL  # optional

3. Deploy:

   npm run deploy
   # or
   vercel --prod

Alternative: Deploy via Vercel Dashboard

1. Connect your GitHub repository to Vercel
2. Set environment variables in the dashboard
3. Deploy automatically on git push

🛠️ Troubleshooting

Common Issues

401/Authentication Error

• Ensure GEMINI_API_KEY is set correctly in .env.local (local) and Vercel dashboard (production)
• Verify your Gemini API key is valid and has proper permissions

Vector Store Not Ready

• Run npm run generate-embeddings to create data/embeddings-gemini.json
• Ensure all required data files exist in the data/ directory

Port Already in Use

• Change the port in dev-server.js or kill the process using port 3000
• Check for other running Node.js processes

Missing Dependencies

• Run npm install to ensure all dependencies are installed
• Check Node.js version (18+ required)

🤖 Demo

Chat with the deployed bot at: https://jgchoti.github.io/

Professional AI that actually understands context and delivers responses that make sense — no generic chatbot nonsense here. ✨