Featured
• Built an agentic AI system in JavaScript on top of a custom harness-engineering pipeline, cutting seat-map creation time from 3 hours to 15 minutes for large-venue events (stadiums, theaters, auditoriums).
• Designed a human-in-the-loop editing studio (Canva-style) to let staff verify and adjust AI-generated seat maps, balancing autonomous generation with accuracy.
• Developed core features for an LLM-powered journalism platform using Next.js, Django, and PostgreSQL, implementing RESTful APIs that handled 10K+ daily requests with under 200ms response time while optimizing database queries to reduce execution time by 40%.
• Engineered responsive UI components with lazy loading and code splitting in Next.js, improving page load times by 40%.
• Built AI-powered content analysis features that processed 500+ articles daily with 92% accuracy.
Featured
Job aggregation system across 500+ company career pages. - cron-based scraping infra - normalized job data across sources - full-stack UI to actually browse jobs → built because job search tools are broken
AirDrop for everybody — even when the two of you are on completely different internets. - streams bytes directly browser-to-browser over an encrypted WebRTC DataChannel (SCTP/DTLS) - a Node.js ws signaling server brokers only the SDP/ICE handshake and never touches your files - custom chunked protocol: binary frame headers, bufferedAmount backpressure, per-chunk SHA-256 + a Merkle-style root - verified cross-network via STUN with a TURN relay fallback → built so file transfer needs no server sitting in the middle
YouTube learning tool that makes content stick. - summaries + quizzes + chat - tracks what you’ve learned - turns passive watching into active recall → built for people who actually want to retain what they watch
AI chatbot for any documentation. Built in ~36 hours after reverse-engineering Kapa. - scrapes + chunks docs - semantic search + context-aware answers - usable UI, not just a demo → built to understand how real AI products are structured
About
I like building products that feel slightly unreasonable for one person. Most of what I know comes from: - picking things that look hard - building them anyway - figuring things out mid-way I care about: speed, product feel, and making things actually work
Skills
taralshah09's coding journey over the past year
Fetching your GitHub activity data
Contributions
Contributed a new shape to Excalidraw end-to-end, from rendering logic to tool integration, merged into the main repo.
If you're building something interesting, let's talk.
