I design systems and ship them to the cloud.

1. 01Wrote a Python pipeline that ingests four provider feeds nightly and normalizes them into a canonical schema
2. 02Built an entity-resolution layer (fuzzy match + manual override table) that unifies player/team IDs across sources
3. 03Modeled the analytics layer in Supabase Postgres with materialized views for hot paths
4. 04Frontend on Next.js (App Router) — hybrid ISR for player/team pages, SSR for live match views, 1,000+ pre-rendered routes
5. 05Deployed on Vercel with CI gates on schema drift

Live at flickstat.com — ~1,200 visitors in month one, 2,000+ and counting now, all organic with zero paid acquisition. Now the foundation for the next product layer.

1. 01Wrapped Meta's Segment Anything Model behind a single-click point/box prompt interface
2. 02Built multi-format export — YOLO, COCO, and Pascal VOC — so it slotted into existing training pipelines without adapter code
3. 03Added batch mode so a folder of images could be processed unattended
4. 04Wrapped it all in a Gradio UI that non-engineers on the team could use directly

Cut manual annotation effort by up to 50% on the team's drone datasets. Now part of the standard preprocessing flow at GarudaUAV.

1. 01Designed a YAML-driven config layer so each pipeline run is fully declarative
2. 02Built a FastAPI backend that orchestrates the five stages with versioned artifacts at each step
3. 03React/TypeScript control UI for kicking off runs, comparing metrics, and inspecting failures
4. 04Dockerized the whole thing — same image runs locally, in CI, and on a cloud worker

Open-sourced on GitHub. Any data source → trained model in one config file, reproducible months later.