Observability tooling, retrieval over real documents, and products I wanted to exist. Some solo, some with a team — each one says which.
Observability for LLM applications. A prompt registry, projects and authentication with email verification, and a dashboard behind feature flags, in an Nx monorepo. The client SDK is a separate package published to npm.

Ask questions about the election process in plain language and get answers grounded in official COMELEC documents rather than in whatever the model remembers. Candidates are browsed as cards and kept to a shortlist. Documents are chunked, embedded and retrieved at question time.
Champion and Best in UI/UX, KomsaiHack 2025

Compose an LLM pipeline by wiring nodes together instead of writing the glue by hand. Chat models, document loaders, embeddings, retrievers, text splitters and vector stores are all nodes with typed ports, so an invalid connection is refused rather than discovered at runtime.
First runner-up, plus Best in Pitching and Best in Q&A — Philippine Startup Challenge 9, regional round

An MVP for an early-stage startup. Every tenant gets its own subdomain or custom domain, resolved per request in middleware before anything renders — so the routing layer, not the app, decides whose data you are looking at. I owned the architecture and built it across the stack.

A calorie and macronutrient tracker for Android and iOS. Built from my own need to track macros without fighting the app doing it.
Contact — 014 / 014