ModelOpt 🚀

AI Model Optimization Engine - Find the perfect AI model for your hardware

ModelOpt helps developers choose practical AI models based on real hardware constraints (GPU, VRAM, RAM), intended use case, and speed-vs-quality preference. It combines deterministic filtering with Gemini-powered recommendation reasoning and installation-ready commands.

Features

• Hardware-aware model recommendations
• AI-powered analysis using Gemini
• Instant Ollama install commands
• Support for 20+ popular models
• Beautiful dark-themed UI
• End-to-end optimization flow: Landing -> Form -> Results

Tech Stack

• Frontend: Next.js 14, TypeScript, Tailwind CSS, ShadCN UI, Framer Motion
• Backend: Vercel Serverless Functions (App Router route handlers)
• AI: Google Gemini API
• Data: Static JSON model + GPU databases
• Validation: Zod + React Hook Form

Installation

Prerequisites

• Node.js 18+
• pnpm (or npm/yarn)
• Gemini API key

Setup

# Clone repo
git clone https://github.com/mohanprasath-dev/modelopt.git
cd modelopt

# Install dependencies
pnpm install

# Set up environment variables
cp .env.example .env.local
# Add your GEMINI_API_KEY

# Run development server
pnpm dev

Open http://localhost:3000 in your browser.

Available Scripts

pnpm dev
pnpm build
pnpm start
pnpm lint

If you prefer npm:

npm run dev
npm run build
npm run start
npm run lint

Project Structure

app/
  api/optimize/route.ts        # Optimization API (validation, ranking, Gemini)
  app/page.tsx                 # Optimization dashboard form
  results/page.tsx             # Results dashboard
  page.tsx                     # Marketing landing page

components/
  results/                     # Results page UI blocks
  ui/                          # ShadCN primitives

lib/
  data/
    models.json               # Curated AI model catalog
    gpus.json                 # Normalized GPU catalog
  utils/
    modelFilter.ts            # Hardware filtering helpers

DEPLOYMENT.md                  # Deployment runbook

How It Works

1. User submits hardware + preferences in /app.
2. API validates payload (Zod), applies rate limiting, and filters/ranks models.
3. Top candidates are sent to Gemini for reasoning and final recommendation narrative.
4. Results page renders primary, secondary, and lightweight recommendations with install tabs.

API Overview

Endpoint:

POST /api/optimize

Request payload:

{
  "gpu": "rtx_4090",
  "ram_gb": 16,
  "vram_gb": 24,
  "use_cases": ["coding", "chat"],
  "speed_preference": 3,
  "deployment": "local"
}

Response includes:

• normalized input
• model recommendations
• summary and warnings
• considered models metadata

Environment Variables

See .env.example for the complete template.

Required:

• GEMINI_API_KEY: Google AI Studio API key for recommendations

Optional:

• NEXT_PUBLIC_APP_URL: App base URL for share links and metadata

Data Sources and Notes

• lib/data/models.json includes researched model metadata (hardware requirements, context, quantization, install commands, strengths/weaknesses).
• lib/data/gpus.json includes normalized NVIDIA, AMD, and Apple silicon entries.
• Citations are embedded as sources arrays and metadata fields for JSON compatibility.

Testing and Validation

Before deploy, run:

pnpm lint
pnpm build

Deployment

See DEPLOYMENT.md for complete Vercel deployment, environment setup, and production checklist.

Troubleshooting

• Missing Gemini key: ensure GEMINI_API_KEY is present in .env.local and in Vercel env settings.
• No results shown: retry optimization flow and verify session storage is enabled.
• Rate limited API: wait one minute and retry (default 10 requests/minute/IP).

License

This project is provided under the repository license in LICENSE.

Built With

• css
• javascript
• typescript