I build the systems behind the work.
APIs. Databases. Integrations. Mobile. I build what happens after the request lands.
0+
Years shipping to production
0+
Enterprise apps, owned start to finish
0+
Daily operations across 17+ warehouses
0%
Uptime. Zero excuses.
Request in. State changes. Outcome out.
Interface
Where the request enters — mobile client, API call, whatever the trigger is.
API & Contract
Payload shape, status codes, and validation — decided before a line of business logic exists.
Business Logic
The rules that decide what a valid request is actually allowed to do.
Database & State
Schema, queries, and the state transitions that outlive any single request.
Integration
Every external service is a dependency with its own failure modes — auth, retries, timeouts.
Observability
Logs and metrics, so a failure leaves evidence instead of a mystery.
Experience
Software for systems that can't pause for a bug.
Software Development Engineer · Falcon Autotech Pvt Ltd · May 2024 – Present
APIs, HHT workflows, warehouse execution — all of it has to stay in sync. I carry it from requirements to the floor.
6+
Full delivery cycle
Requirements to the floor. Mine, start to finish — not just my slice of it.
40%
Less debugging time
Reusable logging frameworks and offline-first design cut the time spent chasing failures.
25–30%
.NET 8 migration
Modularized core services — lower latency, less platform to maintain long-term.
35%
More reliable
Offline-first, defensive design — extended across every warehouse deployment.
The endpoint is only the beginning.
Projects
View all →Deluxe Dhaba — Audio Streaming Platform — 2026 – Present
In progressAudio streaming that fixes itself.
A self-hosted FastAPI resolver behind a Flutter client — built to heal itself mid-playback, not fail silently.
What it does
A stream breaks. It re-resolves itself. Breaks again — it skips and moves on. No dead air, no support ticket.
The trade-off
Self-hosting means I own the uptime. That's the trade: more operational overhead, in exchange for reliability a third-party API was never going to guarantee me.
XVault — Privacy-First Vault App — 2026
ShippedA vault that can't be talked out of it.
A privacy-first vault hardened against modified-APK impersonation — closed with Firebase App Check and gated Cloud Functions.
What it does
A modified APK can fake almost anything about itself. It can't fake a passing App Check attestation.
The trade-off
Gating everything server-side costs latency. It closes an exploit surface client-side checks never could.
Firebase Group Chat Platform
Shipped500 concurrent users is easy. Knowing what each message costs isn't.
Real-time group chat on Firebase, handling 500+ concurrent users at 40% lower read costs through architectural optimization, not brute-force scaling.
What it does
Real-time group chat — messages, presence, and room membership synced live across every connected client on Firebase.
The trade-off
Optimizing read costs meant giving up some of the simplicity of the naive fanout model. Worth it at this scale — the interesting open question is where that trade stops being worth it.
ML Movie Recommender — 2023
ShippedSimilarity beat collaborative filtering by 20%.
A cosine-similarity engine over 5,000+ titles — 20% more accurate than a collaborative-filtering baseline, in my own benchmark.
What it does
Feed it a movie. It finds what's actually similar — cosine similarity, not collaborative filtering.
The trade-off
Collaborative filtering was the explicit comparison point. Why the gap landed where it did for this dataset isn't documented yet — said honestly here, not invented.
Women's SOS Safety App
In progressSOS you can trigger without looking at your phone.
A safety app in progress — gesture-triggered SOS, stealth mode, trusted contacts, live location — on Flutter and Firebase.
What it does
No unlocking. No opening an app under stress. A gesture triggers the SOS — stealth mode, trusted contacts, and live GPS follow from there.
The trade-off
Not documented. Real trade-offs surface once real constraints do, and that hasn't happened yet.
Voice-Enabled AI Chatbot
ShippedI wanted the chatbot to talk back.
A voice-enabled chatbot combining GPT-4 for conversation and DALL-E for image generation — an early, hands-on experiment in multimodal interaction.
What it does
Voice in, conversation out — GPT-4 handling the exchange, DALL-E generating images on request inside the same interaction.
The trade-off
Not documented. This one's here as a real, shipped experiment in stitching a model into a usable product, not because every design decision behind it was recorded at the time.
Engineering labs
View all →Cache — 12 hits · 4 misses
live previewLab 01
Cache Lab
A fixed-capacity LRU cache in front of a database. Tune size, TTL, key popularity — watch it hit, miss, evict, expire.
- Cache-aside strategy — hits, misses, evictions, and TTL expiration
- LRU eviction — tune capacity and watch it decide what to drop
- Deterministic — same seed replays the exact same run, every time
Sortation — 9 delivered · 1 jammed
live previewLab 02
Sortation Lab
A row of sortation lanes under uneven load. Tune capacity and skew — watch queues build, drain, and jam.
- Multiple lanes under configurable load and skew
- Backpressure and jam behavior once a lane hits capacity
- Deterministic — replay any run exactly, byte for byte
Fanout — 25 delivered · 5 dropped
live previewclient 0
client 1
client 2
Lab 03
WebSocket Fanout Lab
One server, several clients, one slow one. Tune buffer size and policy — watch drop-oldest, drop-newest, and disconnect play out differently.
- One server, several clients — one slow enough to fall behind
- Configurable buffer size and drop policy per client
- Deterministic — no randomness at runtime, just a seed
Render tree — naive — 15 of 15 rebuilt
live previewLab 04
Flutter Rendering Lab
A widget tree that rebuilds on every update — scoped or naive. Same tree, same updates, wildly different frame budget.
- Same widget tree, same updates — scoped vs. naive rebuild scope
- Frame budget and jank computed from real per-widget render cost
- Deterministic — the exact same tree, run twice, matches
Open to work
Got something that needs to work?
Reach out about full-stack roles, or anything on this site you want to dig into.
Start a conversation →Open to full-stack roles
mdfarhan422@gmail.com