JahangirBabar.

A 40-lab robotics institute couldn't keep up with its own field, so I built the dashboard that tracks it for them — publications, events, and press across every lab. A Tavily and OpenAI ranking pipeline filters the noise (about 3× better signal), and a Flask scanner with a React review UI handles incremental scans. It cut manual monitoring by roughly 80%.

I owned login and onboarding for a CIAM product millions of people use, where one friction point compounds across every sign-in. By mapping the Okta login journey in Pendo and FullStory and reworking the error prompts with engineering, we cut failed logins about 35% and lifted onboarding completion around 20%.

Backend and ML work on the AWS microservices behind an analytics platform. I added FastAPI alongside Django and reworked the batch-prediction schemas, dropping inference latency from ~500ms to ~50ms, and shipped an LLM content feature that lifted engagement 40%.

I built tooling and CI for a graph-database UI, making a fiddly SQL/Cypher product behave predictably. Rebuilding the pgAdmin components in React and TypeScript pushed feature adoption up 25%, and GitHub Actions pipelines validating Cypher against PostgreSQL cut regression time about in half.