AGRICTOOLS
Turning a catalog-heavy agri business into a tech-first spare parts platform
MISSION BRIEF
India's agricultural machinery market runs on trust, physical catalogs, and phone calls. Dealers identify parts visually, confirm prices through sales reps, and troubleshoot machines through WhatsApp chains. It works — until it doesn't scale.
Agrictools was building toward something different: a unified platform where dealers, technicians, farmers, and manufacturers could discover, evaluate, and transact — without a sales rep in the loop.
I joined as product owner when app adoption was low and the gap between what the product could do and what users actually did was wide.
THE SYSTEM
The platform wasn't one product. It was five surfaces sharing a single backend, catalog, and identity layer:
MARKETPLACE — B2B/B2C buyer app for spares and machines. Multi-seller, role-based pricing, compatibility-aware catalog.
SEVAK — Standalone FOS (field sales) app and light ERP for seller teams. Sales orders, dispatch, CRM-lite — all on the same source of truth as the marketplace.
AGRICTRADE — Lead-gen and ads platform for high-value machine discovery. Enquiry-only listings, verified pricing for KYC'd manufacturers, subscription-based leads.
CATALOG ENGINE — The backbone. 50k+ SKUs, machine compatibility graph (Type → Brand → Model → Variant), OEM/OES governance, role-aware pricing rules.
AI LAYER — RAG-based troubleshooting assistant trained on internal video knowledge. Automated ingestion pipeline. Spare part recommendations from diagnosis.
MY ROLE
Product Owner — but the scope went beyond documentation.
I owned end-to-end product decisions across discovery, commerce, and AI. I also prototyped features myself when speed mattered more than polish, worked directly with the engineering team on system design, and contributed to the design system that kept five surfaces visually consistent.
The work that mattered most wasn't writing PRDs — it was figuring out how to change user behavior without asking users to change.
WHAT I SHIPPED
Three areas worth going deep on:
01 — CATALOG SCANNER
Bridge between offline habits and digital ordering
Dealers trusted their physical catalogs. Typing SKUs one by one into an app was never going to win. So I built a different entry point — one that met them exactly where they already were.
Prototyped → iterated → shipped with the team. Order creation time dropped from ~30 minutes to 3–5 minutes.
02 — AI TROUBLESHOOTING ASSISTANT
RAG system trained on internal video knowledge
Every troubleshooting call was the same tribal knowledge being rediscovered. I designed and helped build a RAG pipeline that ingested Agrictools' video library — auto-transcription, chunking, embedding, vector storage — and turned it into an interactive AI assistant. Set up using Langchain, custom pipelines, and a continuous ingestion loop so it got smarter as new content was uploaded.
Case study in progress.
03 — MARKETPLACE PLATFORM
Multi-surface system design from the ground up
Designing the rules that govern a marketplace is harder than designing its screens. Pricing model across five buyer roles, catalog governance separating OEM from OES, channel exclusivity logic, Sevak as a parallel FOS layer — all had to work together from a single source of truth.
Case study in progress.
STACK & TOOLS USED ON THIS PROJECT
Langchain RAG / Vector DBs OCR pipelines Figma Flutter (design handoff) GCP Razorpay WATI Surepass KYC API