ai-gateway/implementation_plan.md

Storyline AI Gateway - Implementation Plan

Project Overview

A FastAPI-based gateway for e-learning modules (Articulate Storyline) to access LLM services (Gemini, OpenAI) with centralized authentication and rate limiting.

Tech Stack

• Framework: FastAPI
• Server: Uvicorn
• Rate Limiting: Slowapi (limiter)
• Auth: Header-based API Key (X-API-Key)
• LLMs: Google Gemini, OpenAI

Directory Structure

• app/main.py: Application entry point and middleware configuration.
• app/core/: Configuration and utilities (limiter, settings).
• app/api/deps.py: Shared dependencies (authentication).
• app/api/router.py: API versioning and route aggregation.
• app/api/endpoints/:
  • storyline.py: Generic endpoint.
  • gemini.py: Dedicated Gemini endpoint.
  • openai.py: Dedicated OpenAI endpoint.

Configuration

Managed via .env file:

• API_KEY: Secret key for Storyline modules.
• GOOGLE_API_KEY: API key for Google Generative AI.
• OPENAI_API_KEY: API key for OpenAI.
• PORT: Server port (default 8000).

API Endpoints

All endpoints are versioned under /api/v1.

1. Gemini Chat

• URL: /api/v1/gemini/chat
• Method: POST
• Headers: X-API-Key: <your_key>
• Body: {"prompt": "string", "context": "string"}

2. OpenAI Chat

• URL: /api/v1/openai/chat
• Method: POST
• Headers: X-API-Key: <your_key>
• Body: {"prompt": "string", "context": "string"}

Rate Limiting

• Applied globally/per endpoint: 20 calls per minute.

Future Steps

• Add logging (WandB or file-based).
• Implement response caching.
• Add more LLM providers (Anthropic, etc.).