Chef Jul

The MetLife Hackathon 2025 challenge: build something meaningful in 48 hours. Our team chose personalized nutrition — a space where generic meal plans fail users because they ignore individual preferences, restrictions, and goals.

The core technical challenge was latency. Generating a full weekly meal plan with per-meal nutrition data required multiple sequential LLM calls — a naive implementation would take minutes, killing the interactive feel we needed to win.

Team Lead and Full-Stack Developer. Set the technical direction in hour one, assigned work streams, and kept the team unblocked across 48 hours. Personally architected and implemented the parallelized LLM pipeline and the Split-Brain multi-agent workflow.

Instead of sequentially generating each meal, the pipeline fans out: all seven days of meal generation run in parallel via Promise.all, with each day's meals generated concurrently within that batch. This alone cut end-to-end generation time by 40% to under 55 seconds on constrained hackathon hardware.

User preference adjustments use a Split-Brain pattern: a lightweight preference agent holds shared state and responds to changes in ~2 seconds by regenerating only the affected meals rather than the full plan.

2nd place finish at MetLife Hackathon 2025. Judges highlighted the real-time preference adjustment experience as the standout differentiator.

Shipped a production-ready Firebase backend with auth, Firestore persistence, and a React frontend — all within the 48-hour window.