Rauan Khamza, computer science at CityU Hong Kong, from Astana. Agent infrastructure at Gensmart; Python teaching at CityU.
An MCP server that gives AI agents an interface to LinkedIn for posting and scheduling: a REST API, a scheduling database and per-user agent tokens. The Next.js dashboard reaches the API through a server-side proxy, so its service token never reaches the browser.
// what an agent calls
create_post({
text: "Shipped the scheduler today…",
schedule_at: "2026-10-27T09:00+08:00",
draft: true,
})
→ { id: "post_…", status: "scheduled" }Dockerized MCP servers that let my agents do the parts of student life that sit behind university logins. Each one drives a headless browser on Cloudflare Browser Run, signed in through CityU's single sign-on, and all of them are reachable from Claude Code, claude.ai and my phone.
University emails go through a pipeline that spots events with a sign-up form. I get a yes-or-no question in Telegram; on yes, the browser fills the form from what it knows about me and sends back a screenshot. Nothing is submitted until I confirm, and every correction I reply with is remembered for the next form.
An ESP32 with a servo that flips the air-conditioning wall switch in my room, controlled from a web page or my phone. Commands reach the board over a WebSocket held by a hibernating Durable Object, so the switch reacts at once and costs nothing while it waits.
A serverless service that syncs event guest lists from Google Sheets into a SQL store. Edited rows update in place, bad rows wait in a queue without stalling the rest, and guests removed from the sheet are flagged. Alongside it, a CRUD app for event management on MongoDB JSON schemas.
Exam practice for CityU courses, open to every student. Sign-in is a passkey, Learn mode walks through the lecture slides, and students can upload their own question banks and share them. Most questions come from real tutorials, labs and past quizzes.
A Jupyter extension on PyPI. It runs hidden tests on each cell, and when the code fails, an LLM asks a spoken Socratic question instead of giving the answer. It became the base of the tutorial notebooks for CS1302A, the new AI-assisted version of CityU's intro Python course, where students guide an AI to rebuild it.
$ pip install socratic-watchdog
# In [3]: a cell fails its hidden test
# the tutor asks instead of fixing it:
# "What does your loop do on the last item?"As a PALSI leader I give peer instruction and live technical help in CS1302, a 430-student team-based Python course. This semester I am a teaching assistant for CS1302 and CS1302A.
I practise on Codeforces, and a problem only counts once the same solution is accepted in C++, Python and Java. These figures load live from codeforces.com.
| Year | What | Where | Result |
|---|---|---|---|
| 2026 | DIVE into Agentic Coding: Can You Teach AI to Teach You? | IEEE TALE 2026, with C. Chan and six co-authors | Accepted, full paper |
| Agentic Coding grant, AI coursework integration | City University of Hong Kong | One of 5 picked from 20 | |
| socratic-watchdog | PyPI | Published | |
| drillit.app | CityU students | 33 sign-ups in its first 10 days |
I grew up in Astana and finished school at Nazarbayev Intellectual School. After a first year of computer science at Nazarbayev University, I moved to Hong Kong to continue at CityU.
Most of what I build sits between AI agents and the systems they need to reach. I run a lot of my own life that way: my timetable, deadlines, printing and gym log are MCP servers my agents can use.
The other half is learning. I teach Python, helped build an AI-assisted version of CityU's intro course, and run an exam-practice app that other students use every day.
I speak Kazakh, Russian and English, and I am looking for internships in AI and backend engineering in Hong Kong.
Rauan KhamzaSoftware engineer, Hong Kong
Internships, hackathon teams and hard problems in AI or backend work are all welcome.
