Ashesh Srivastava

Things I've shipped

Every product here was designed, built, and shipped by me — solo. Each page tells you what I built, who it's for, what I decided and rejected, how the AI agents work, how I measured quality, and what it cost me to learn. GitHub links included, because "trust me" isn't a source.

Start with the case studies

🔦Betting against the models — and losing wellQuicSnap → Shutter Labs — the flagship case study🎨Building on the model that killed my startupCanvas — the image workspace built after
Orbit
Personal feed ranker that ingests 800+ sources I actually follow and turns each morning into one ranked 7am digest — X noise dropped, videos demoted but never deleted, and one tap away from a Claude voice conversation. Dogfooded daily for 42 days.
Agentic architectureClaude agentsRankingRAGPythonSQLite804sources → one daily digest
blip
News as vertical reels: an iOS app with karaoke captions over generated narration. Tuned the caption pipeline from 58% to 92% usable output.
iOSLLM pipeline92%usable output (from 58%)
JobFairy
A job-search copilot for any candidate — verified H-1B sponsor filter, agent research on every role, tailored resumes, agent-filled applications, and outreach that lands in inboxes.
Browser agentsNext.js$0.02vs $1.14 per task
Astrape
A usage model for a Texas bill-credit electricity plan, validated to within $0.17 of the real bill — and a 999 vs 1,000 kWh paradox worth the read.
ModelingEnergy$0.17model vs real bill
Marketplace automation
Built the custom platform and ~35-person ops team behind 77 US marketplace stores (QEG Automation).
EcommerceOps at scale$593Kpeak monthly GMV
Ecommerce consumer brands
Built two private-label consumer brands — A Baby Cherry® and Decor & More® — sourced from factories in China, Vietnam, and India, sold on Amazon.
DTCAmazonPrivate Label