AI Systems
MCP servers that expose a real platform to AI assistants, agent pipelines that do real editorial work behind human approval gates, and the plumbing that keeps them in production. Built as systems, authenticated, scoped and audited, never as demos.
AI Engineering - Agents, MCP Servers and Durable AI Workflows
The platform around the model that takes AI features from demo to production: agents that plan, act and verify, and MCP servers over a business's own systems. Around them sit routing across model providers, durable workflows for long jobs, memory scoped per client, and guardrails that keep a person on every write. The live MCP server behind this site is the working example, and the same engineering ran on an agent-first AI workflow platform serving more than a thousand tenants. It runs on self-hosted infrastructure, so nothing leaves the building.
Md Moniruzzaman
Browse This Work by Tag
Other Areas I Build In
iOS and Android apps on React Native and Expo, built as a surface of the platform behind them: the same backend, auth and data model as the web, released through the stores on one cadence. The shelf runs from a booking platform and job-sharing network for UK minibus and coach operators to a traveller help app delivered over eighteen months, a life skills app for students, a local sports app, a distributor's billing tool and a fan platform.
Desktop software in Electron and React for the work that stays on a machine: offline-ready, packaged for Windows, macOS and Linux, and wired to the platform's backend like any other client when there is one. Keyboard Tracker, an employee activity tracker with its own report generator, is the example on this shelf.
Web platforms and the sites in front of them, on React and Next.js against a headless CMS: admin dashboards that run a business day to day, customer portals, a cooperative's ledger and loans, and public sites that load fast and read well to search engines and AI crawlers. Server-rendered, localized and measured, whatever the size.
The infrastructure and release work that keeps a platform live: a production migration off managed cloud onto self-hosted Kubernetes with no recorded downtime, Dockerized services behind Cloudflare, CI/CD that ships on push, and mobile releases taken through store review. Backups, TLS and monitoring are part of the job, not extras.
Embeddable pieces that carry a platform onto someone else's website. The Prebooked widget lets a minibus or coach operator put booking on their own site, served by the Fully Booked platform behind it, so customers book without leaving the page. Small to install, and held to the same standard as the product it extends.
The pieces around the platforms that are neither an app nor a site: gm-libs, the reusable TypeScript packages for auth, data, interfaces and background work that every Webanion frontend and backend starts from, and the design trail of a product, from an interactive Figma prototype and its UI/UX to the visual assets and tutorial videos that took it to market.