I architect AI systems
that survive contact
with production.
Technical Lead and Software Architect with 10+ years shipping distributed, cloud-native systems across HealthTech, startups and consultancies. I turn fuzzy AI ambitions into services with owners, budgets, latency targets and a page someone can wake up to. Everything on this domain — the work, the writing, the code, the CV — is one place, kept current.
How I work
Architecture first
Domain boundaries, event contracts and failure modes on paper before a line of glue code — DDD, EDA, saga patterns and state machines. Most AI projects die from missing seams, not missing models.
Spec-driven, not ad hoc
Plan → implement → independent review before human sign-off. Trading ad hoc prompting for a protocol-based lifecycle lifted my startup team's throughput by roughly 40%, with a human accountable for every merge.
Own the operations
Infra, CI/CD gates, telemetry and on-call. Replaced unreviewed daily pushes with a gated weekly release train and OpenTelemetry observability — now about two issues a month are caught before they reach users. I build things I'm willing to be paged for at 3am.
Selected work
All projects →
AI-powered personal finance app: track income, expenses and goals, with an AI coach that reads your real data. Serverless AWS backend for near-zero idle cost, built through the same plan → build → review pipeline I run for clients.
Featured writing
All writing →
I Hired a Robot Dev Team. My Architect Billed Me €40 Before Lunch.
A solo dev's field notes on building a three-persona AI pipeline — architect, builder, reviewer — and the very human moment I brought it all back home for a quarter of the cost.
Stack I reach for
TypeScript · JavaScript · Node.js · NestJS · React · Vue · React Native · Angular · Go · Python · FastAPI · C# (.NET)
LangGraph · LangChain · Multi-agent orchestration · spec-driven development · RAG · LLM integration · Claude · OpenAI · GitHub Copilot
CDK · Lambda · API Gateway · Cognito · DynamoDB · Step Functions · EventBridge · SES · S3 · CloudWatch · EC2 · CloudFront · KMS · WAF · Docker · GitHub Actions · OpenTelemetry
PostgreSQL · MongoDB · DynamoDB · Redis · Kafka · RabbitMQ · SignalR · GraphQL
Apps I build and run myself.
Small number, fully owned: infra, auth, mail, on-call. Each one is a real product with real users, not a demo.
AI-powered personal finance tracker and coach. React Native app on a serverless Go/AWS backend — Cognito, API Gateway, Lambda, DynamoDB — with OpenAI-driven spending insights and EventBridge-scheduled alerts over SES. Running in a closed beta right now — get in touch if you'd like to test it.
A weekly time-tracking app: log hours against categories — sleep, work, food, study, gym, commute — and see the week as a ring chart out of 168 hours, with a running tally of what's still unaccounted for. Running in a closed beta right now — get in touch if you'd like to test it.
Project three
Slot held for the next app.
More slots reserved — this list grows one entry at a time.
Usman Ali Siddiqui
Technical Lead — distributed systems and agentic AI, 10+ years across HealthTech, startups and consultancies.
Experience
Technical Lead (Head of Engineering)
Took technical ownership of a live, contractor-built platform (100+ daily users across Europe and the UK) with no tests, environments, backups or monitoring, and closed the business-critical risks within two weeks. Replaced unreviewed daily pushes with a gated weekly release train — isolated staging, mandatory review, CI test and coverage gates, OpenTelemetry/Dash0 observability. Decoupled job ingestion and backups from the monolith, grew the engineering team from zero, and rolled out my multi-agent orchestrator as the team workflow, lifting throughput by ~40%. Part-time, alongside a full-time role.
Senior Software Engineer · Acting Technical Lead
Lead a 6-engineer team modernising Widex's hearing-aid fitting software to the cloud — a platform used by ~20,000 hearing-care professionals and handling 700k+ sessions a week. Shape the event-driven architecture for extending it to new device platforms: domain boundaries, data harmonisation and integration-event contracts, approved through R&D architecture review. Built custom spec-driven AI skills encoding our domain and architecture conventions, now the default workflow for 10+ engineers and adopted by sibling teams; automated dependency updates with Renovate (~1 day → ~2 hours each).
Senior Software Engineer
Took features from refinement to production across domain-driven microservices (C#/.NET, React), backed by unit, contract and system tests. Built cross-domain data flows with GDPR-compliant pseudonymisation of medical data, and contributed to 7 production releases, including cloud-to-device communication. Mentored 3 engineers who have since grown into senior roles.
Full Stack Developer
Built the frontend for a B2B application in React and TypeScript against a C# (.NET Core) backend, and an internal ticketing desktop app with React and Electron.
Research Contributor, EU Dementia Care Project
Part-time research held concurrently with Master's studies: built a Python computer-vision application for motion tracking and prediction of hand-movement patterns, improving automated patient-monitoring accuracy.
Lead Software Engineer
Led a team of 4 building a greenfield service-marketplace platform (TypeScript, React, Node.js, MongoDB, AWS EC2/S3), owning architecture across backend, frontend and mobile. Designed a pub-sub architecture for real-time carpooling communication and cut third-party API costs with Redis-based caching.
Full Stack Software Engineer
Delivered scalable B2B/B2C applications for international clients using TypeScript, React, Node.js, Angular, MongoDB and GraphQL, collaborating with cross-functional and offshore teams.
Software Engineer
Developed and deployed hardware-integrated enterprise systems — biometric attendance, canteen billing, visitor management — for HBL Bank and Dolmen Group, using C#, SQL Server, RFID and Angular. The software still serves roughly 10,000 users daily.
Notes from building the thing.
Published on Medium, indexed here so it doesn't get lost.
I Hired a Robot Dev Team. My Architect Billed Me €40 Before Lunch.
Field notes on building a three-persona AI dev pipeline on cloud APIs, hitting a €40-per-feature bill, and moving it all local for roughly a quarter of the cost.
EventBridge or Step Functions? I Built Both to Find Out
Building the same workflow two ways to see where each AWS orchestration tool actually earns its keep.
I Built a RAG-Powered Smart PDF Assistant — Here's What I Learned
Upload a PDF, ask questions in plain English, get grounded answers with sources — and what broke along the way.
Learning RAG the Local Way: Build a Free In-Memory Retriever + Classifier
A from-scratch retriever and classifier that runs entirely in memory, no hosted vector database required.
My First Micro LLM in Python: Build a Mini RAG System in Minutes
A minimal RAG pipeline in Python, built to understand embeddings and retrieval before reaching for a framework.
Building a Peer-to-Peer Rental Platform with Event-Driven Microservices
An open-source NestJS and Kafka boilerplate for peer-to-peer rental platforms.
Go Serverless Microservices: A Clean Architecture Boilerplate for Go Developers
A production-ready Go boilerplate for AWS serverless microservices, structured around Clean Architecture.
K-Means and Finding Centroids
Notes on K-Means clustering and centroid-finding from the neural-network-interpretability research behind my Master's thesis.
Where are my Hands?
Computer vision for hand and object detection in images.
Code you can read.
Libraries and tools pulled out of production and research work.
Upload any PDF, ask questions in plain English, and get grounded answers with confidence scores and source page references.
Event-driven architecture with EventBridge and saga orchestration of long-running workflows using AWS Step Functions.
Event-driven microservices boilerplate for peer-to-peer rental platforms, built on NestJS and Kafka.
Production-ready Go boilerplate for AWS serverless microservices with Clean Architecture.
Minimal in-memory LLM and RAG pipeline demonstrating embeddings, vector search and context-aware querying in a few lines of Python.
Open to lead and architect roles.
Based in Germany, open to remote or relocation. Best over email — I answer everything that isn't a template.