Agentic AI & Frontend Developer
I'm Abdullah Shaikh, a full stack agentic AI developer. I ship agents that do real work, from a live AI app that helps teachers plan lessons to autonomous loops that fix failing tests on their own. My stack is Python, FastAPI, the OpenAI Agents SDK, Claude Code and Next.js.
01Selected work
Agents that do real work, and the interfaces that make them usable. Each one ships with source or a live demo.
01
A local-first autonomous agent that triages Gmail, drafts replies and LinkedIn posts, and routes work through skills — with Claude Code as the reasoning engine, Obsidian as the dashboard, and a human approval step before anything is sent.
02
An interactive CLI that supervises Docker Compose projects with OpenCode — it waits for Docker, optionally rebuilds containers, then checks their health every three minutes and writes a running report to Markdown.
03
Helps teachers create class notes, tidy messy notes, and turn them into revision summaries and lesson plans, using skill-based AI agents with real-time streaming and saved history.
04
Free, exam-style practice questions for the official Panaversity certification quizzes — built by students, for students, so the real quiz has no surprises.
A furniture marketplace built for the GIAIC Marketplace Builder Hackathon 2025, with products managed in Sanity and cart state in Redux.
A content-driven blog powered by Sanity as a headless CMS, with Clerk for authentication.
Small command-line utilities published to the npm registry.
A task list app with add, delete, clear-all, and a theme switcher.
02Toolkit
What I reach for when building interfaces, agents, and the pipeline between them.
Responsive, accessible interfaces backed by headless content.
Tool-using agents, autonomous loops, and the APIs that serve them.
The workflow that ships it.
03About
I'm Abdullah Shaikh, an Agentic AI and Frontend Web Developer based in Karachi.
On the frontend I work with Next.js, TypeScript, and Tailwind CSS, often backed by a headless CMS like Sanity. On the AI side I build with Python and the OpenAI Agents SDK, and increasingly with coding agents such as Claude Code and OpenCode — wiring them into watchers, loops, and MCP servers so they can take on real, repetitive work.
What I care about most is keeping a human in the loop: agents that handle the routine and leave the final decision to a person, behind interfaces that are clear enough to trust. The goal is simple — software that looks good and solves real problems.
04Contact
I'd love to hear about it — whether it's a web app, an AI agent, or both. Send a message or reach out on any of these.