Software Engineer at ABS Courier · Cairo, Egypt
Backend engineer who makes systems hold up at scale.
For two years I've been building the backend of ABS Courier's new shipment platform with NestJS, PostgreSQL and Redis. I tune queries on tables with tens of millions of rows, keep transactions and lock windows short, and moved millions of legacy rows into production while the old system kept running.
ABS Courier
I'm on the backend team for the company's new shipment platform: the system clients, couriers and employees use every day to create, move and track shipments. I've shipped 600+ commits across shipments, fleet, ticketing and reporting. Most of my work is in the parts that have to stay fast and correct under load: the database, the background jobs, and the migration off the legacy platform.
I was the primary developer of the NestJS/TypeScript service that migrated millions of rows from the legacy SQL Server system into the new PostgreSQL production database, then kept the two in sync through Change Data Capture.
SQL Server
Legacy platform
CDC sync engine
NestJS · runs every 30 min
lsn 0x0004A3F1 · seq 3Arabic/English district resolver · phone & text normalization
Syncing · batch 1
PostgreSQL
New production system
(LSN, seqval) cursor it stores in PostgreSQL,
applies them in foreign-key order in bounded batches, then saves the new cursor. During peak hours it stands
down so the live legacy database isn't put under extra load.
Shipments, accounts, pickups, call-center records, tickets and client users moved with checkpointed resume, automatic retries and SMS alerts on failure. Bounded batches let the sync catch up large backlogs without long-running transactions.
I replaced row-by-row writes with set-based statements into a shipment table range-partitioned on
(awb, created_at). Preloaded lookup caches removed the N+1 queries.
INSERT INTO shipment
(awb, created_at, …)
VALUES (…), (…)
ON CONFLICT (awb, created_at)
DO UPDATE SET
status = EXCLUDED.status;A district resolver tries exact names, then a 450+ entry curated map, then fuzzy matching. It runs alongside Arabic text and Egyptian phone-number normalization.
A few of the problems I've tracked down and fixed on the production platform.
Report subqueries were sorting shipment history row by row. A partial, sort-ordered index lets PostgreSQL read the rows already in order, so the per-row sort becomes an index seek.
CREATE INDEX … ON shipment_history
(awb, audit_date DESC)
WHERE reason_id IS NOT NULL;Sortper row, inside every subqueryIndex seekrows read already in orderThe yearly wallet report ran 36 sequential queries. I rewrote it as one grouped query that scans only the
requested year's partitions, and batched per-row updates into one UPDATE … FROM (VALUES …) per 500 rows.
Timeouts during bulk imports traced back to connection-pool starvation: five workers holding long transactions
on a 20-connection pool. Validation now runs outside the write transaction, in short chunks under
lock_timeout with per-AWB advisory locks.
FOR UPDATECSV exports run on Bull queues, read with keyset pagination and stream to disk with backpressure, so memory stays flat at any size. Files are zipped to private storage behind 24-hour signed URLs.
Manual and generated ranges, sequential and random alphanumeric formats, and safe claiming under concurrent load. Many workers can ask for numbers at once and no number is handed out twice.
NDJSON uploads run in queues with row-level validation, live progress and automatic client notification on failure. Redis execution locks keep a retried job from writing twice.
Cars, drivers, missions, trip logs and driver-to-driver handover, plus the backend for the fleet mobile app.
Most of the internal ticketing system: filtering, search, sorting, pagination and ticket analytics.
International shipment flows and app-wide timezone support, including timezone-aware date-range filtering.
I built the shared end-to-end test setup and its GitHub Actions pipeline, with smoke, regression and authorization suites covering every endpoint. I also added a tenant guard that limits client writes to their own sub-accounts, and ran API security testing across the platform.
Each square is a day of commits I pushed to ABS Courier repositories, taken from GitHub. Merge commits are left out.
Graduated with a B.Sc. in Computer Science from Cairo University.
Joined ABS Courier's backend team.
First modules on the new client platform: sub-accounts, shipments, pickups and business settings.
Started the legacy SQL Server → PostgreSQL migration and the CDC sync service.
Moved bulk shipment import onto queues; added delivery verification and logging.
Shipped the AWB registry and allocation system.
Busiest month: shipment reporting, rerouting, and automated notifications when a bulk import fails.
Optimized reports; the wallet report went from 36 queries to one.
App-wide timezone support and a new payment gateway integration.
Performance pass across every client dashboard endpoint.
Whether it's a role, a project, or a stubborn Postgres query, my inbox is open.