Scaling Multi-Tenant B2B E-Commerce and Credit Risk Underwriting
Growing engineering 3.3x (15 to 50+) while replacing legacy monoliths with a proprietary multi-tenant WMS and credit engine.
01 The Business Situation
Urban Harvest operates a B2B enterprise perishables and agricultural commerce platform with multi-tenant warehouse operations and high-volume trade credit. Business was rapidly growing from ~₹40 Cr to ~₹70 Cr monthly.
02 What Was Broken
Legacy Node.js and C# monoliths suffered from high database locks, slow warehouse pick-pack cycles, and reliance on expensive third-party SaaS vendors (Zoho, ElasticRun).
The B2B trade-credit business had ~₹10 Cr in outstanding credit with projections reaching ~₹100 Cr, but lacked algorithmic underwriting—sales teams made credit decisions manually.
Data pipelines were fragmented, leaving inventory forecasts and warehouse order prediction reliant on outdated T+1 spreadsheets.
03 What I Decided & Why
Scaled engineering organization 3.3x (15 to 50+ engineers) in 6 months
Established engineering managers and tech leads for dedicated pods, implemented high-bar interview calibration, and took full ownership of the technology budget reporting to the CEO.
Built an algorithmic credit risk model for B2B trade credit
Automated prepaid, partial, and full-credit decisions based on real-time trade history and repayment velocity, protecting capital while supporting business expansion.
Migrated legacy monoliths to Go and Java microservices on Kubernetes
Eliminated two third-party SaaS vendors by launching a proprietary multi-tenant WMS with an immutable Cassandra audit ledger.
Rebuilt data warehouse with Trino, Superset, and Airflow
Collapsed business intelligence reporting from T+1 days to real-time minutes, powering machine-learning order prediction for warehouse stocking.
Engineered an AST-based AI Context Graph across Golang repositories
Cut LLM token consumption by 60%+ during AI-assisted code generation and automated log troubleshooting.
04 Measurable Results
05 Lessons for Startup Founders
Never allow manual trade-credit decisions in B2B commerce. High sales numbers mean nothing if credit defaults blow up your working capital.
Third-party warehouse SaaS software quickly becomes a tax on unit margins once volume hits scale. A tailored internal WMS pays for itself in months.
Scaling engineering from 15 to 50 requires rigorous calibration and pods. Without clear pod ownership, hiring more developers only slows down shipping.
06 Technologies Used
WORK WITH UMESH GUPTA
Ready to scale your architecture
without expensive trial and error?
Book a 20-minute call to evaluate your engineering bottlenecks. Response: Reply the same business day (IST).