Commerce Engineering
An honest breakdown of my commerce engineering knowledge — what I've built in production, what I'm actively learning, and what's on my roadmap. No inflation.
Level definitions
Commerce System Map
Every eCommerce system is a pipeline. This is the full lifecycle — from product discovery through to post-purchase analytics — and where I am in understanding each stage.
Supporting Systems
Technical Stack
How the technical stack is organized across my projects — from browser to database.
React 19
Server and client components
Next.js 16 (App Router)
GharBazaar + this portfolio
TanStack Start
SSR + server functions (VedaGlow)
TypeScript 5
Strict typing across every project
Next.js Route Handlers
59 endpoints in GharBazaar
TanStack server functions
Typed server calls with auth middleware
Zod
Runtime schema validation
Supabase Auth
Session and identity management
PostgreSQL
Primary database in both products
Supabase
Managed Postgres, Auth and Storage
Row-Level Security
155 policies across both products
PostGIS
Geospatial proximity (GharBazaar)
Cart Engine
Persistence, validation, totals
Order Management
Six-state lifecycle
Coupon Engine
Discount rules + usage caps
Razorpay
Server-side signature verification
Vitest + Playwright
196 unit + 23 e2e specs
GitHub Actions
Lint, test, build, e2e on every push
Vercel
Deployment + scheduled jobs
Git + GitHub
Version control
Core Commerce
The fundamental systems that every eCommerce platform must have
Product listings, variants, categories, images, SEO-friendly URLs, and schema.org structured data.
Add to cart, quantity management, persistent cart state across sessions, and cart totals with discount application.
Multi-step checkout covering address collection, order summary, payment selection, and order confirmation.
Full order lifecycle — order placement, status tracking, admin order views, and customer order history.
Discount codes, percentage and flat discounts, validity windows, minimum order rules, and single-use enforcement.
Product, order, and user management with role-based access, secure authentication, and operational views.
Payments & Auth
Payment workflows and user authentication systems
Razorpay integration covering order creation, checkout, server-side verification and failure handling — plus a split advance-payment and cash-on-delivery path built for Indian D2C.
Payment state is never taken on trust from the browser — the signature is verified on the server before an order is treated as paid.
Supabase Auth manages customer and admin identity in both products. This portfolio additionally runs a hand-built JWT flow, so I have implemented both the managed and the from-scratch version.
Multi-role system with permission-based route and action gating — separating buyers, sellers, and platform admins.
Automated refund processing and financial reconciliation on top of a live payment integration.
The refunded payment state already exists in the VedaGlow schema; automating the refund path itself is the next step.
Marketplace Systems
Multi-sided platforms — buyers, sellers, and platform operators
Listing lifecycle, location-based discovery, university-based student housing search, and multi-role workflows.
Implemented across 58 tables and 59 API routes, and deployed on Vercel with the database connected. Pre-launch: seller onboarding is still in progress.
Vendor isolation, separate catalogs, order routing to correct vendor, commission calculation, and payout systems.
Data & Search
Product discovery, search, and inventory systems
PostgreSQL schema design across a commerce store and a marketplace platform — 68 tables and 155 Row-Level Security policies in total.
Structured, validated, and authenticated REST APIs with consistent response shapes and pagination.
Stock tracking, variant-level inventory, low-stock alerts, and reservation during checkout.
Filtered property discovery with saved searches and a nightly job that matches new listings against them.
Built for property discovery. Product search over a large catalog — relevance tuning, synonyms, typo tolerance — is a different problem I have not yet solved.
AI & Performance
AI-powered features and engineering performance practices
Code splitting, lazy loading, image optimization, caching, and Core Web Vitals engineering.
Natural-language product search, semantic recommendations, review summarization, and RAG-based product Q&A.
Vector database, embedding generation, and retrieval-augmented generation for commerce queries.
Advanced Commerce
Order operations, returns, and marketplace financials
First-party event capture across the listing funnel, plus platform performance monitoring.
The event capture is built and the data model supports funnel analysis. Turning it into admin-facing dashboards is still ahead.
Engineering Insights
Specific lessons from VedaGlow and GharBazaar — things I didn't understand until I built them.
A payment confirmation arriving from client-side code proves nothing — anything the browser sends can be forged. Verifying the Razorpay signature on the server before marking an order paid is the line between a payment system and a hopeful one. Building that in VedaGlow changed how I think about every client-supplied value.
Card-first checkout underperforms in Indian D2C. Cash on delivery converts, but exposes the seller to refused deliveries. Splitting the difference — a small advance captured online, balance on delivery — took real design work and did more for conversion than any UI change I made.
With four roles and 58 tables, enforcing access only in route handlers means every new endpoint is a fresh chance to leak someone's KYC document. Moving the rules into 128 Row-Level Security policies changed the failure mode: a forgotten filter now returns nothing instead of returning another user's data.
Campus distance is the entire product promise, so it cannot be something a seller types in. Making it a PostGIS value computed by a trigger — and recomputed whenever a property's pin moves — removed a whole class of drift bugs that application-side calculation would have kept reintroducing.
Persistence across refreshes, price changes while an item sits in the cart, coupon eligibility recalculated on every mutation — each is a distinct edge case. Persisting locally keeps guest checkout frictionless, but only because totals and line items are re-validated server-side before an order exists.
Several Row-Level Security gaps in GharBazaar surfaced because a test asserted a user could not read something and it turned out they could. Writing the tests alongside the features caught design problems, not just regressions.
Engineering Philosophy
Commerce systems need to work reliably under real-world conditions — incomplete data, edge cases, user errors. I prioritize correctness and resilience over technical elegance.
There's a lot I'm still learning. The most dangerous engineer is one who overstates their experience. I clearly separate what I've shipped, what I'm studying, and what's planned.
eCommerce is a full-stack problem. UX, API design, data modeling, and deployment all affect the outcome. I care about every layer because they're all part of the user's experience.
Great eCommerce is not about features — it's about systems. Catalog, cart, checkout, payments, orders, and inventory are all interconnected. Understanding how they compose is what separates a developer from a commerce engineer.
The best way to evaluate my commerce engineering is to read the actual case studies — architectural decisions, challenges, and tradeoffs.