AI Full Stack Engineer Intern at Legal-Jarvis (2026-04 – Present)
Legal Maia — AI Legal Document Automation Platform
- Architected a production multi-agent AI system using Google ADK to automate fact extraction, case analysis, question generation, and drafting across 213+ petition templates.
- Designed agent orchestration and session memory through a whiteboard architecture, coordinating specialized agents and preserving drafting context across multi-step reasoning pipelines.
- Built tool-calling and structured-output generation, integrating external REST APIs and multiple LLM providers (OpenRouter, Azure OpenAI, Gemini) for reliable, production-grade document generation.
- Developed a RAG document-retrieval pipeline using Pinecone and Qwen embedding models to semantically retrieve relevant legal documents, improving drafting accuracy and grounding.
- Led production deployment on Google Cloud Run using Docker, Cloud Build, and Artifact Registry, with intelligent petition selection served via Supabase.
- Owned AgentOps and the production lifecycle — monitoring, validation, pipeline debugging, and iterative prompt and agent optimization to improve system reliability.
