I build the systems a business actually runs on.

A fleet telematics SaaS I build solo: any GPS hardware, or a driver’s phone, in — a daily verdict for the fleet owner out.

Fleets across Egypt and the Gulf already own cheap GPS trackers and are stuck on weak local portals. LemursEye takes whatever hardware they have: a Traccar bridge decodes 200+ device protocols, and everything downstream runs on a Kafka event pipeline, so the platform never cares which vendor sent a position. Quarkus workers enrich each position, reconstruct trips from cached and offline data, colour history by motion, evaluate geofences and raise alerts with a mandatory cooldown so nobody gets spammed.

History lives in TimescaleDB under per-tenant row-level security; Keycloak handles sign-in, roles, and staff-versus-customer separation with audited impersonation. A Next.js console and a Flutter app, which doubles as a phone tracker, show vehicles live over WebSocket. Instead of a map to stare at, the owner gets a morning digest of what happened, framed in money. It is Arabic-first and right-to-left, with a public site in 12 languages, and ships as Quarkus native images on a hardened VPS behind Caddy with CI/CD.

Solo founder, live at lemurseye.com and taking pilot requests. I built it end to end: backend, web console, mobile app, infrastructure, brand and the bilingual installer docs. Every major choice is written down as an architecture decision with the alternatives it beat, and that is the part I would most like to talk through.

tracker protocols

200+

languages

12

commits in 2 months

150+

engineer

1

A bilingual research platform that refuses to publish a number without its source.

Orphan-care statistics are scattered, contradictory and usually undated. The platform publishes indicators for 166 countries under one rule: no figure goes public without its source, its year and a confidence grade from A to E. Where no acceptable number exists, the gap itself is published as a declared gap rather than left blank.

Partner organisations submit figures that move through a seven-stage review with four roles, per-item decisions and an append-only audit log. Nothing is overwritten, so an editor can always answer where a published number came from and who let it through.

countries

166

published figures

1,400+

review stages

7

roles

4

Inventory and accounting for steel and cement distribution, replacing a business run entirely on spreadsheets.

One sale has to post to inventory, the customer ledger, the driver account and the bank book together or not at all. That logic lives in 52 Postgres functions rather than in application code, so a dropped connection cannot leave half a sale behind.

Access is enforced by 79 row-level-security policies in the database, including hiding profit margins from cashiers. Nothing is ever edited or deleted: corrections are linked entries and the original stays visible. It works offline in the warehouse and replays writes on reconnect. Entirely in Arabic, right to left.

Postgres functions

52

RLS policies

79

destructive edits

0

A multi-tenant WhatsApp Business CRM, grown from an early-stage codebase to a production system.

I joined a young codebase as the main backend developer and took it to 37 Postgres tables and around 217 REST endpoints. Campaigns run through a Redis and BullMQ queue with per-message delivery tracking, so a tenant sending thousands of messages can still see exactly which ones landed.

WhatsApp Flows are RSA-encrypted, inbound webhooks are HMAC-verified, and a Gemini assistant answers from each tenant’s own uploaded documents with their API keys encrypted at rest.

tables

37

endpoints

217

Rota, timesheets and payroll for a UK care provider, built as the first phase of the product.

Care coordinators were running rotas, leave and expense approvals across spreadsheets and phone calls. The dashboard puts the whole cycle in one place: recurring shifts per client, a month rota view, timesheets against a payroll cut-off, and expense claims that a coordinator approves or queries in line.

I built the first phase over three months and handed it on: the Next.js dashboard, the Cloud Functions behind it, and the Flutter app the assistants use in the field.