Software Engineer | Data Scientist

Marine researchers survey up to 50 km of reef a day and then label the footage frame by frame. Omarine turns that into a review job instead of a transcription job. I built the ingest pipeline pushing transect imagery into GCS at 30 fps, the Postgres schema tying frames to predictions and human-validated labels, and semantic search with cross-frame object grouping so researchers query by visual concept, not filename. An evaluation dashboard surfaces detection counts, confidence distributions, and mask-area ratios. Estimated 20–30% reduction in labelling time.

A literature review usually lives in twelve browser tabs, a notes doc, and a citation manager that agrees with neither. Annota is one workspace instead: projects, PDFs, saved links, citations, comparison tables, search. Content is extracted from PDFs and web pages, then annotated through the OpenAI API so every summary stays anchored to the passage it came from. TipTap-based editor with citation chips, table editing, word counts, and sanitised HTML storage.

Budidayaplus is a web app that helps Indonesian fish farmers track what actually affects a harvest. Farmers log validated water-quality readings, fish and feed sampling, and daily tasks against a production cycle for each pond. We built it as a team, with a Django Ninja API over Postgres and a Next.js frontend in TypeScript and Tailwind. Starting a cycle fires a Django signal that turns day-of-culture templates into dated, assignable tasks, so nobody has to build the schedule by hand. We worked test-first throughout and shipped it with Docker.

Surveys and interviews first, code second. Students didn’t need another social feed — they needed a reason to talk to someone in week one. The prototype covers interest-based matching, guided icebreakers, community discovery, and anonymous chat for students not ready to post under their own name.