Md. Abu Ammar · Backend & AI Systems Engineer
I build and run Partners.com.bd — a live marketplace and business-social platform across Bangladesh and the UK: its .NET API, Flutter apps, and legacy MVC→API migration. Before that, at Masjid Solutions, wallet and ACH payments on a platform moving millions of dollars a year and AI-powered kiosk monitoring across 200+ devices on Azure. MS CS researcher in quantum machine learning.My undergraduate thesis explored quantum machine learning — variational circuits and encoding methods on PennyLane simulators — and the research thread continues through my MS: Bangla POS tagging with knowledge distillation, multi-output CNNs, ensemble methods. I bring a production engineer's discipline to research code.I own the .NET API, Flutter apps, and MVC→API migration behind Partners.com.bd — a live marketplace across Bangladesh and the UK. Before that at Masjid Solutions: the Apple Pay / Google Pay and Authorize.Net ACH stack on rails moving millions a year for 20,000+ users, Azure Vision AI across 200+ kiosks, and 7–8 production releases a week through CI/CD I architected — DDD and vertical slices throughout. The hero beside this text is a quantum classifier I wrote from scratch; drag the data points.
This is a real 2-qubit variational quantum classifier, training live in your browser. The RY boxes are rotation gates whose angles θ are learned parameters; the vertical line is a CNOT entangling gate; the dials update by gradient descent using the parameter-shift rule. Drag the two data points and it retrains on the new problem. Every number comes from a 166-line, unit-tested quantum simulator — learn how it works, from zero.
$MM+
annual payment volume supported
200+
kiosks monitored across 60+ orgs
20,000+
users served by payment systems
350+
production deployments per year
featured system
A serverless visual-QA platform that watches a fleet of donation kiosks: it screenshots every device on a schedule, runs Azure AI Vision over each frame, and — when it finds a fault — restarts the kiosk and alerts the responsible operator, all without anyone standing in front of the screen. Built as a .NET 9 Azure Functions app, orchestrated by .NET Aspire, deployed to Azure Container Apps via `azd`.
background-job-system · 2026-07
Designed and built the background-job subsystem for the Partners.com.bd marketplace backend (.NET, ASP.NET Core, SQL Server) — a dedicated Hangfire worker with an atomic-enqueue outbox, a dead-letter queue, idempotent handlers, least-privilege SQL isolation, and OpenTelemetry-based observability. The single place all async work now lives.
[dotnet][hangfire][distributed-systems][reliability]
auth-architecture · 2026-07
Designed the authentication architecture for the Partners.com.bd marketplace (ASP.NET Core Identity, MediatR, SQL Server) around three principles: one core decides *who you are*, each client edge gets the session artifact safest for its medium (HttpOnly cookie for the server-rendered web, bearer JWT + rotating refresh for native mobile), and every session — web or mobile — must be revocable and enumerable server-side. Architected across two accepted ADRs after a four-stream R&D review: one decision core, two transport edges, revocable sessions on both, and a JWKS issuer boundary — with the single-core convergence of the user logins and a full Critical/High hardening set under 832 tests.
[dotnet][security][identity][oauth]
payments-platform · 2026-06
Designed and built the payments platform for the Partners.com.bd marketplace backend (.NET, ASP.NET Core, SQL Server) — one generic `dbo.Payment` money ledger that serves every paid feature, a vendor-agnostic `IPaymentGateway` abstraction with a country-aware resolver, a single self-healing settlement coordinator, and a Hangfire reconciliation backstop. It replaces a legacy wide table (`dbo.OnlinePayment`) that stored money as text and trusted the client's callback.
[dotnet][payments][distributed-systems][reliability]
ammar2022bangla · Directed research (CSE498), North South University · 2022
[nlp][bangla][bert]
Part-of-speech tagging for Bangla — a low-resource language whose main benchmark, Microsoft IL-POST, is severely class-imbalanced — using contextual embeddings from three Bangla BERT models. On macro-F1, which weights all 30 classes equally, a neural network beats a decision tree overall but wins by feasting on the majority classes; the tree scores lower yet spreads its errors more evenly across the tagset. That asymmetry motivates an inverted distillation: read the class counts in the tree's leaf nodes as a probability distribution and distill that "dark knowledge" from the tree into the neural student — a de-biasing signal rather than a compression trick.
ammar2022quantum · B.Sc. thesis (CSE499), North South University · 2022
[quantum-ml][quanvolution][cvqnn]
A comprehensive review of the state of the art in quantum machine learning, paired with hands-on classification experiments: two first-generation hybrid quantum-classical models — a quanvolutional neural network on a gate-based simulator and a continuous-variable quantum neural network on a photonic simulator — trained on MNIST and compared head-to-head against classical baselines of comparable size on accuracy and convergence. Both hybrids trail their baselines; the margins, and where each model loses, are the contribution.
I work where production engineering meets machine learning research — connecting business operations, software architecture, AI, payments, and cloud infrastructure into systems that run every day.
Today I lead Web & App Development at Partners Online (Bangladesh & UK), the company behind Partners.com.bd. I design and run the platform end to end — a .NET API, the Flutter apps, and a legacy MVC→API migration — deploying constantly and designing with DDD, vertical slice, and clean architecture principles so the systems stay healthy long after they ship.
Before that, at Masjid Solutions, I built systems across their full lifecycle: on a payment platform supporting millions of dollars in annual volume for 20,000+ users, the Apple Pay / Google Pay wallet stack and Authorize.Net ACH billing end to end, plus the operations layer around them — reporting, alerting, automated failed-payment alerts, and scheduled donation summaries; KioskVisionAI, which watches 200+ donation kiosks across 60+ U.S. organizations with Azure Vision AI; and an automated Salesforce synchronization platform that eliminated manual CRM entry. Every product line shipped through CI/CD I architected — 7–8 zero-downtime production deployments a week.
The research thread runs in parallel: my undergraduate thesis explored quantum machine learning — variational circuits, encoding methods, hybrid classical-quantum models on PennyLane simulators — and continues through my MS: Bangla POS tagging with knowledge distillation, multi-output CNNs, ensemble methods.
My mission: continuously improve systems, automate the repetitive, and innovate at scale.