ai lab · entity online
His memories, his own corner of the internet · running · web, llm
The problem: He always wanted his son's memories stored in a customized fashion over the internet.
The build: Vibe-coded the entire website in Google AI Studio — multiple pages, each customized around one little boy.
Family expenses by voice or photo — even via iOS Shortcuts · running · voice, vision, mobile
The problem: A real pain point: no single app to track family expenses and project investments.
The build: A Claude-built PWA — iOS and Android both — expenses added by voice or image, Gemini reading them at the backend. iOS Shortcuts log a spend without even opening the app.
Fast Food Destroyer · running · llm, mobile
The problem: A junk-food habit that needed a distraction — and an incentive to stop.
The build: A Claude-built PWA with small gamification and real incentives. Starve the gremlin.
An Android app sized for very small hands · side project · mobile
The problem: Hand a toddler a tab, and all he wants is to tap the screen.
The build: So that became the whole app — an Android build out of Google AI Studio: tap anywhere, delight follows.
BharatPe's first AI feature in production · shipped · vision, llm
The problem: Chargebacks and disputes were handled manually — high turnaround, unhappy merchants.
The build: A chargeback agent that auto-resolves disputes — won or lost — within seconds. BharatPe's first AI in production, hand-tuned from 75% false positives to 2%.
Awaaz — "Sunta Hai, Samajhta Hai" · running · audio, llm, automation
The problem: Tens of thousands of calls flow through voice AI and human call centers — unheard at scale.
The build: Sunta hai, samajhta hai. Churn, sentiment, and 15+ more signals built from every conversation — no PRD, shipped in two weeks.
Vision models running shop photo verification end to end · shipped · vision, automation
The problem: Shop photo verification was completely manual.
The build: An image analytics agent automating the entire validation — plus new signals about every merchant no reviewer had time to see.
3rd-party vendors + an in-house orchestrator · running · voice, automation, llm
The problem: Human-based calling was never scalable — it always ended up taking a back seat.
The build: Partnered with multiple voice AI companies to bring AI to 20+ use cases across BharatPe — while building the in-house, in-app stack on live LLMs.
SQL writer · banner creator · PPT creator · running · sql, image-gen, automation
The problem: Recurring dependencies kept eating the team's time — SQL, banners, decks.
The build: A mini agent for each: SQL Writer, Banner Creator, Presentation Creator — built with whatever tool fit: Google AI Studio, Cursor, NotebookLM.