system online — v13.7
senior software engineer :: mobile · backend · ml
I design and ship production systems end-to-end — from Flutter and React Native clients through Node/NestJS and Laravel services, down to the ML pipelines behind them. 50+ apps, 80+ web platforms and three published research papers.
/services
One accountable engineer for the whole journey — no agency layers, no hand-offs. You talk to the person who actually builds it.
Your product on the App Store and Google Play — designed, built and launched end-to-end.
The online store, booking site or admin panel your business runs on — fast and reliable.
Practical artificial intelligence inside your product — grounded in published research, not hype.
/work
Three of the 50+ products I've shipped — anonymized out of respect for the businesses behind them, with the real numbers kept.
consumer services · national scale
A consumer platform where people book and pay for an essential service. Built end-to-end: booking flows that keep working on weak networks, integrated payments, and updates shipped without a minute of downtime.
telecom · self-service
A telecom companion app where customers manage their account, pay bills and browse media-heavy services — turning support calls into taps.
marketplace · two-sided
Customers on one side, hundreds of independent providers on the other — matching, scheduling and tracking in a single product, with tools for both sides.
/process
A simple, transparent path from idea to launch — you'll always know where your project stands.
Free and no-obligation. You describe the idea; I tell you honestly what it takes — including when the answer is "you don't need an app for this."
A clear scope, timeline and quote in plain language — you'll know exactly what you're getting before anything is built.
You see real, working progress at regular checkpoints — and can change course early, while changes are still cheap.
App-store submission, hosting and monitoring — plus updates and support after launch, so it keeps working as your business grows.
/architecture
Not a list of buzzwords — a running topology. Every subsystem below has shipped to production. Click a node to inspect it.
/systems
Individual products come and go — the fleet numbers don't. Thirteen years of shipping, measured in production deployments, not side projects.
Fifty-plus production apps shipped to the App Store and Google Play — marketplaces, telecom self-service, commerce, booking platforms and enterprise tools. Full lifecycle every time: UI/UX, API integration, offline tolerance, release trains, store operations.
largest single deployment: 300K+ installs · 10K+ daily actives
Eighty-plus websites and admin dashboards — the control planes behind the apps. Enterprise dashboards, e-commerce storefronts, CMS builds and internal ops tooling, each backed by APIs I own end-to-end.
a dashboard is a product too — ops teams are users
/research
The ai.pipeline subsystem, peer-reviewed — machine and deep learning work that shipped to journals, not just to notebooks.
“Assembly Line Anomaly Detection and Root Cause Analysis Using Machine Learning” — a 7-model unsupervised ensemble (HBOS · IForest · KNN · CBLOF · OCSVM · LOF · ABOD) detecting manufacturing anomalies and tracing them to root cause.
one detector lies — seven detectors vote
“Hybrid Deep Learning for Mental Workload Classification Using EEG with Enhanced Preprocessing and Interpretability” — classifying mental workload from EEG signals, with preprocessing built for noisy biosignals and interpretability treated as a requirement, not an afterthought.
a model you can't explain is a model you can't ship
“Toward Intelligent Cognitive Load Estimation: A Systematic Synthesis of Machine and Deep Learning Techniques, Challenges, and Prospective Developments” — mapping how ML and DL estimate cognitive load: techniques, open challenges, and where the field goes next.
reading the whole field so the next model starts smarter
/code
Three snippets, three layers — the patterns I actually ship: composition roots, swappable providers, and ensembles that vote.
// Composition root — features depend on abstractions, never on each other.
Future<void> bootstrap({required Flavor flavor}) async {
WidgetsFlutterBinding.ensureInitialized();
final env = Environment.resolve(flavor);
final di = ServiceLocator.instance
..registerLazySingleton<ApiClient>(() => DioApiClient(env.baseUrl))
..registerLazySingleton<AuthRepository>(
() => AuthRepositoryImpl(remote: di(), cache: di()))
..registerFactory<OrdersBloc>(() => OrdersBloc(di()));
// Crashes are a data source, not an embarrassment.
FlutterError.onError = CrashReporter.record;
runZonedGuarded(
() => runApp(App(env: env)),
CrashReporter.recordZoneError,
);
}// Module boundaries mirror domain boundaries — not folder aesthetics.
@Module({
imports: [
TypeOrmModule.forFeature([Order, OrderItem]),
BullModule.registerQueue({ name: 'order-events' }),
],
controllers: [OrdersController],
providers: [
OrdersService,
// Payments are an interface; Stripe is an implementation detail.
{ provide: PAYMENT_GATEWAY, useClass: StripeGateway },
OrderSagaOrchestrator,
],
exports: [OrdersService], // nothing else leaks out of this domain
})
export class OrdersModule {}# IEEE Access — assembly-line anomaly detection, 7-model ensemble.
# One detector lies. Seven detectors vote.
DETECTORS = {
"hbos": HBOS(contamination=0.02),
"iforest": IForest(n_estimators=200),
"knn": KNN(method="mean"),
"cblof": CBLOF(),
"ocsvm": OCSVM(kernel="rbf"),
"lof": LOF(n_neighbors=32),
"abod": ABOD(),
}
def score(window: np.ndarray) -> AnomalyReport:
votes = {name: d.predict(window) for name, d in DETECTORS.items()}
consensus = majority_vote(votes, quorum=4)
return AnomalyReport(
flagged=consensus.any(),
root_cause=trace_root_cause(window, consensus),
)/principles
Seniority isn't knowing more frameworks — it's making trade-offs explicit. These are the ones I make on repeat.
/log
Fourteen years of deployments, upgrades and scale events — tail -f of a career.
/contact
No contact form theater — a protocol. Fill the payload, dispatch the request, get a real reply.
“dispatch” simply opens your own email app with the message pre-filled — no signup, no tracking.