Jack of Clubs
Tech
Built the software the business runs on.
Six months as founding engineer: twelve projects shipped and live, taking a Telegram-and-Excel operation across twenty collection centres onto software built for it.
Hands played
Bullion
₹125 Cr+ a yearThe procurement system of record, from the collection centre to the farmer's bank: purchase orders, inwards with farmer KYC, an immutable stock ledger, push notifications at every step. Farmer payment that took days now clears the same day.
One inward: identity, land record, grade, weight, approval 
Every centre's day on one screen Bull Connect
Every farmer, rememberedAn in-house farmer CRM that replaced a paid tool, built around how the business sells: every call logged, each farmer's stock, last offer and expected price on record, so the telecaller rings the moment the market hits his price.
A farmer record as the calling team sees it Bull Compass
Every castor village, mappedA live map of castor area and trade volume across 22 APMC territories, built by matching nine lakh hectares of government crop data to village polygons. It decides where the next centre opens.
Where the crop is 
Where Bull already trades Bull Logistic
One record per truckEvery truck from centre to factory as one record: driver and vehicle papers, multi-stop pickups, weighbridge photos and weight reconciliation at the factory, push notifications at every stage.
One trip, six states KOSHI's poster engine
Reads the pack, draws the poster, checks its own workThe Python engine behind KOSHI's posters: it reads a product pack, plans the Gujarati text, draws the card with an image model, reads the result back with OCR and repairs what is wrong. Nothing leaves unverified.
The AI lead pipeline
Hundreds of leads a dayField supervisors write one Gujarati line in Telegram; Gemini turns it into a structured lead and pushes it to the CRM. No data entry for ninety supervisors across twenty centres.
The data groundwork
The master data under everythingEighteen and a half thousand Gujarat villages scraped from the state land-records portal as the reference for every project; eighty-nine thousand farmer records audited; an offline app so supervisors can capture leads from villages with no signal.
BISAG-N, summer 2025
A farmer assistant on a CPUFine-tuned Llama 3.1 8B on 2.5 million Kisan Call Centre records into an assistant that answers farmers in Indian languages in under three seconds on an 8 GB machine with no GPU.
Reliance Jamnagar, summer 2024
Nine sites on one screenA server-health dashboard across nine manufacturing sites, SQL automation for the data behind it, and a model forecasting gasifier plant life.