LLM Code Deployer

Automated code generation and deployment system using LLMs (via aipipe.org) and GitHub Pages.

Features

• ✅ Generate HTML/CSS/JS apps using OpenAI GPT-4o-mini via aipipe.org
• ✅ Automatic GitHub repository creation
• ✅ Automatic GitHub Pages deployment
• ✅ Clean HTML extraction (removes markdown and explanations)
• ✅ Windows-compatible file paths
• ✅ TLS certificate handling for Python 3.13

Setup

1. Install Dependencies

# From project root with venv activated
cd D:\tds-proj1\llm-code-deployer\backend
pip install -r requirements.txt

2. Configure Environment Variables

Edit backend/.env:

OPENAI_API_KEY=your_aipipe_token_here
GITHUB_USERNAME=your_github_username
GITHUB_TOKEN=your_github_personal_access_token
STUDENT_SECRET=your_secret_key

Important:

• Get aipipe token from: https://aipipe.org/login
• Generate GitHub token: https://github.com/settings/tokens (with repo scope)
• No inline comments in .env file!

3. Fix CA Certificate Issues (Windows Only)

If you see TLS/SSL errors, clear these environment variables:

$env:CURL_CA_BUNDLE=$null
$env:SSL_CERT_FILE=$null
$env:REQUESTS_CA_BUNDLE=$null

Running the Server

Option 1: Manual Start

cd D:\tds-proj1\llm-code-deployer\backend

# Clear CA env vars (important!)
$env:CURL_CA_BUNDLE=$null
$env:SSL_CERT_FILE=$null
$env:REQUESTS_CA_BUNDLE=$null

# Activate venv
D:\tds-proj1\.venv\Scripts\Activate.ps1

# Start server
python -m uvicorn main:app --reload --host 127.0.0.1 --port 8000

Option 2: Use Start Script

.\start_server.ps1

API Usage

Build & Deploy Endpoint

POST http://127.0.0.1:8000/build
Content-Type: application/json

{
  "student": "username",
  "brief": "Description of the app to generate",
  "secret": "your_secret_key"
}

PowerShell Example:

$env:CURL_CA_BUNDLE=$null
$env:SSL_CERT_FILE=$null
$env:REQUESTS_CA_BUNDLE=$null

Invoke-RestMethod -Uri "http://127.0.0.1:8000/build" -Method POST `
  -Headers @{ "Content-Type" = "application/json" } `
  -Body '{"student":"myname","brief":"A todo list app","secret":"rajsecret2151"}' `
  -TimeoutSec 180

Response:

{
  "status": "deployed",
  "repo": "myname-app",
  "pages_url": "https://your-username.github.io/myname-app/"
}

Testing

Use the provided test script:

.\test_build.ps1

Project Structure

llm-code-deployer/
├── backend/
│   ├── .env                 # Environment variables
│   ├── main.py             # FastAPI application
│   ├── generator.py        # LLM code generation
│   ├── deploy_repo.py      # GitHub deployment
│   └── requirements.txt    # Python dependencies
├── frontend/
│   └── index.html          # Frontend UI
├── test_build.ps1          # Test script
└── README.md              # This file

Key Updates

1. Fixed API Endpoint

• ❌ Old: https://api.aipipe.org (doesn't exist)
• ✅ New: https://aipipe.org/openrouter/v1/chat/completions

2. Improved HTML Extraction

• Removes markdown code blocks (html...)
• Extracts only HTML content
• Handles LLM explanations

3. GitHub Pages Auto-Enable

• Automatically enables Pages after deployment
• Uses main branch as source

4. Better Error Handling

• Validates GitHub credentials
• Handles existing repositories
• Force push to update content

5. Windows Compatibility

• Uses tempfile.gettempdir() instead of /tmp
• Handles CA certificate bundle issues

Troubleshooting

Server won't start

# Make sure you're in backend folder
cd D:\tds-proj1\llm-code-deployer\backend

# Check Python path
python --version  # Should be 3.13

# Reinstall uvicorn
pip install uvicorn --force-reinstall

TLS Certificate Errors

# Clear bad environment variables
$env:CURL_CA_BUNDLE=$null
$env:SSL_CERT_FILE=$null
$env:REQUESTS_CA_BUNDLE=$null

# Reinstall certifi
pip install certifi --upgrade

GitHub Authentication Fails

1. Verify token at: https://github.com/settings/tokens
2. Ensure token has repo scope
3. Generate new token if expired
4. Update .env file (no comments!)

Generated HTML is wrapped in code blocks

The updated generator.py now:

• Uses stricter prompts
• Extracts HTML using regex patterns
• Removes markdown formatting

License

MIT

Credits

• OpenAI GPT models via aipipe.org
• GitHub API via PyGithub
• FastAPI framework