What MealPe is

MealPe is an institutional food-services SaaS for universities, workplaces, hostels and industrial campuses in India. It covers multi-vendor food court ordering, counter billing, mess meal plans with subscription billing, vendor settlement, institutional admin dashboards and attendance hardware. I am the sole owner of its backend, PostgreSQL schema, AWS infrastructure and deployment pipeline, and have been since April 2024.

My role: Sole backend, database, DevOps and server owner at MealPe Online Food Ordering POS Pvt. Ltd. (Apr 2024 – present).

The problem it solves

Institutional food service looks like ordinary food ordering until you run it. Most of a day's orders land in one short lunch window, so the system has to be sized for a rush rather than for an average. Money flows three ways, between the diner, the institution (which often subsidises meals) and many independent vendors who each expect to be paid exactly what they earned. Every institution configures its own vendors, menus, meal plans and staff roles, and counter hardware such as biometric terminals sits on campus networks that drop out. All of this had to keep working while the user base grew twentyfold, with one backend engineer.

How it works

The server is the authority on prices, permissions and state, and the client is never trusted to create an order: a confirmed payment does. Work that does not need to happen inside a request, such as notifications, settlements and reporting, goes through a transactional outbox in the same Postgres database, so it is durable without adding a message broker. Meal quotas are pre-computed so a checkout reads one value instead of scanning history, which is what lets the lunch rush through. Vendor payouts are split at payment time and reconciled rather than synced. Per-institution access is a permission set enforced in middleware rather than a policy engine. Scaling was mostly database work: composite indexes on real access patterns, pagination, connection pooling and moving side work out of the request path.

The outcome

The platform grew from roughly 1,000 to over 20,000 registered users while I owned it, and now handles over 70,000 meal transactions a month and around 120,000 API requests a day. I cut the AWS bill by about 60% without downgrading the service and moved releases to a zero-downtime PM2 and NGINX pipeline. I also handle client onboarding, technical demos and production incidents. Most of my technical writing comes straight from this system.

Technology used

  • Node.js
  • PostgreSQL
  • Supabase
  • AWS EC2
  • AWS S3
  • NGINX
  • PM2
  • Socket.IO
  • Razorpay

Engineering highlights

  • Scaled from ~1,000 to 20,000+ registered users with zero planned downtime, as the only backend engineer
  • Cut the AWS bill by ~60% without downgrading service: logging overhead, S3 lifecycle rules, static assets moved to NGINX
  • Made payment confirmation, not the client callback, the event that creates an order
  • Pre-computed meal quotas so checkout stays O(1) through a lunch rush that brings 80% of the day's orders in 90 minutes
  • Split-payment vendor payouts that pay each vendor exactly once, even when the customer's share is a few paise
  • Durable background jobs on a Postgres transactional outbox instead of adding Kafka
  • Zero-downtime releases with PM2 cluster mode, graceful reloads and NGINX