AI-LUMA

A chat assistant that routes each request to whichever specialist agent and model provider suits it. The backend is five separate services behind an API Gateway rather than one process, because the agent work and the billing work fail in completely different ways and should not share a restart.

What was built

• Architected a distributed, multi-tiered microservices backend of 5 services: Gateway, Auth, Agent, Chat and Billing, with an API Gateway and Redis rate limiting.
• Built a LangGraph and LangChain orchestration engine where a supervisor agent coordinates 4 specialist agents for coding, vision, document parsing and web search.
• Integrated 4 model providers, Google Gemini, DeepSeek, Groq and OpenRouter, so each request can be routed on cost, latency and quality rather than being pinned to one vendor.
• Wrote tests for the service endpoints and checked code quality across the 5 service boundaries.
• Built the React frontend with an artifacts panel, Firebase authentication and Razorpay subscriptions.