Four products,
built solo,
without an engineer.

FMCG distribution in India is run through a network of plants, carrying-and-forwarding agents, and distributors. A planner managing 200+ SKUs across 20+ CFAs answers three questions every morning: Which SKUs are running out, where? Which orders need to go out today? Are the buffers set right?

Most mid-market companies answer these through ERP exports, Excel pivots, and planner intuition. When a consultant leaves, the methodology leaves with them. StockSmart replaces this with a daily-run analytical engine grounded in Theory of Constraints buffer logic — five-band availability, FMS classification, demand-driven replenishment, monthly buffer revisions with planner sign-off.

The position is deliberate: lighter than enterprise SCM, deeper than DMS, persistent unlike consulting deliverables. Multi-tenant architecture with database-level Row-Level Security (penetration-tested), pg_notify audit trail, daily automated backups, comprehensive test coverage. Engine validated against real FMCG distribution data; in channel partner discussions with two TOC consultancies.

A Chrome extension that sits next to any YouTube video and tells you, in fifteen seconds, whether it's worth watching — and whether the central claims hold up. Built for the audience I belong to: Indian analysts, journalists, investors who burn three hours a day on long-form video and want a faster way to triage it.

Three tabs. Summary gives a verdict and breaks the video into themes with a note on why each one matters. Reality Check verifies the video's claims against the live web and flags contradictions, with healthy / unverifiable / contradicted states. Dig Deeper opens a conversation with the video — pull threads, compare to other sources, get answers grounded in the transcript.

The interesting engineering problem wasn't the AI — it was getting the cost per video down low enough that a ₹499/month price could carry it. Currently distributed via private Chrome Web Store link to early users.

Most garment factories I worked with at Goldratt run on WhatsApp, scribbled bin cards, and an Excel sheet that the owner's son maintains. The big ERPs (SAP, Tally) are either too expensive or too generic. The cheap apps are toys.

Stitchwise sits in the middle. Production scheduling, inventory tracking, order management, and a stage-wise leakage report — because in this industry, you don't lose margin from one big problem, you lose it in slow drips at cutting, sewing, QC, and packing. Built for the UP/Haryana SME cluster specifically, in the language and workflows they already use. Target price ₹30,000/month, which lands between "free Excel" and "₹2 lakh SAP implementation."

Roughly 88% feature-complete. Pre-revenue. The next step is the first paying customer, not the next feature.

A full-stack productivity app I built for myself — task capture, note-taking, project planning, an LLM that has context on all of it. Genuinely useful, used daily, not for sale.

The reason it's here is the cost-engineering work. Naive AI apps run a single big model on every request and burn money. Vidur classifies each request and routes it — trivial calls go to Claude Haiku, reasoning goes to Sonnet, and everything cacheable is cached at the Cloudflare edge. The result was an operating cost reduction from $1.30/day to ~₹130/month, roughly 97%.

The point isn't the app. The point is that this is the kind of cost discipline AI deployments need at enterprise scale — and that the operations frameworks I've used for inventory work surprisingly well on token spend.