Aaditya Kushwaha

Abstract schematic of a distributed backend

What I actually do

Most of my work is the unglamorous half — the part that decides whether the demo survives contact with real traffic.

I've spent four years shipping production systems end to end: the FastAPI service, the Next.js console that operates it, the Flutter app that consumes it, and the network it runs on. Not four separate jobs — one system, seen from four sides.

The through-line is enforcement at the right layer. Tenant isolation belongs in the database, not in a code review checklist. Compliance rules belong in a type-checked default-deny, not a wiki page. A circuit breaker belongs around every external call, because the third party will time out at 3am.

I write about the failures because they're the part worth reading — the outage that taught me ElastiCache cluster mode is not Redis, and what it cost to find out.

→Building agentic infrastructure tooling at Altacee
→Writing up production postmortems worth reading
→Open to remote roles — 4+ hours EU / US-East overlap

Selected systems

Four of them

Abstract diagram — Ride-hailing platform

Abstract diagram — School-transport safety

Abstract diagram — The network it runs on

01

Aerovahan

2025 — 2026

Ride-hailing platform

Dispatch and matching for a live marketplace: driver-offer caps, radius expansion, dynamic surge, hot-zone broadcast, demand forecasting, and a rider–driver bid flow. I owned the backend, the admin console, and both Flutter apps.

—863 commits
—94 routers
—110 services
—98 migrations
—~135k lines

Diagnosed a production outage caused by ElastiCache cluster-mode semantics — cross-slot deletes, blocked KEYS scans, and a single Uvicorn worker taking health checks down with it.

FastAPI · SQLAlchemy 2 async · Postgres · Redis · Next.js · Flutter

02

Aero Squad

2026

School-transport safety

Parents watch a bus in real time; schools verify arrivals with a TOTP gate check that works with no signal. Geofenced trips over OSRM map-matching, and a KYC pipeline doing DigiLocker lookups, OCR and face-to-document matching behind circuit breakers.

—372 commits
—22 screens
—0 → shipped in 10 weeks

Every external API sits behind a breaker, because a KYC provider timing out should slow one onboarding, not the fleet.

FastAPI · PostGIS · OSRM · Flutter · Riverpod · arq

03

Blitzlearning

2026

Multi-tenant AI LMS

Tenant isolation enforced by Postgres row-level security rather than application code — the database refuses cross-tenant reads even when the app is wrong. SCORM, cmi5 and AICC delivery, white-labelled per tenant.

—415 commits
—RLS at the database
—pgvector + Meilisearch

Isolation you can test: the integration suite asserts a tenant cannot read another's rows even with a forged tenant id.

FastAPI · Postgres 16 · pgvector · Next.js 14 · Turborepo · uv

04

Altacee

2024 — now

The network it runs on

Designed and operate the estate end to end: dual-WAN with failover, 802.1Q segmentation across three isolated networks, RSTP bridges on redundant NIC pairs. Internal services have no public listener at all — reachable only across the mesh.

—17 interfaces
—2 Proxmox estates
—18 VMs
—~65-node tailnet

Provisioning, firewall changes and DNS run through an MCP server as typed, audited operations rather than console clicks.

OPNsense · Proxmox · Docker Swarm · Traefik · Tailscale · Cloudflare

Open source

Repos I committed to

Writing

All postmortems →

Things that broke in production, and what the fix actually was.

Most of my work is the unglamorous half — the part that decides whether the demo survives contact with real traffic.