For AI agents
Hamza publishes a human.json: what they're open to, on what terms, and how to approach them. Agents read the file before reaching out.
https://humanjson.org/u/hamza/human.json
About
I'm the founder of ZenML, an open-source AI platform that takes teams from ML pipelines to agents. What I think about most is getting agents and ML systems to work in production: evals, orchestration, and what breaks once real users arrive. If you're working on the same problems, I'd like to hear about them.
Open to
Conversations with people working on agents in production, evals, or machine learning in production. Bring a real problem you're working on.
- Topics: agents in production, evals, ML in production, LLMOps, MLOps.
Projects
One AI platform from pipelines to agents. Open source, Python.
5,600+ GitHub stars and 665 forks (October 2026).
- Kitaru2026
Agent traces you can run, not just read: durable execution, checkpoints and replay for AI agents.
310 GitHub stars in its first months.
- ZenML MCP server2025
Connects MCP clients like Claude Desktop and Cursor to ZenML pipelines.
An example of LLM document-processing pipelines, built on ZenML.
- cdtm-mcp2026
An MCP server for the CDTM Community Portal: search CDTM students and alumni from Claude in plain language.
How to move agents from demos to reliable production systems: harnesses, checkpointing, retries, replay, and where Kitaru fits.
An agent is a while loop that talks to a model, calls a tool and writes to a file system. What durable execution does and does not guarantee.
The architecture of ZenML: how one framework bridges local development and production on Kubernetes.
Panel on running ML in production: best practices, implementation challenges and MLOps trends.
Taught part of Data Science Dojo's LLM Bootcamp.
ZenML's podcast on ML in production. I co-hosted six episodes, including an interview with Ines Montani of spaCy.
Indie hacking, and running an MLOps startup in Germany.
Community AMA about ZenML and MLOps.
Skills
- Python●●●●●ZenML, cdtm-mcp
- MLOps●●●●●ZenML, Local Code, K8s Scale: Building a Vendor-Agnostic AI Platform, Navigating the MLOps landscape (panel)
- LLMOps●●●●●ZenML, ZenML MCP server, LLM Bootcamp (Data Science Dojo)
- Agents●●●●●Kitaru, ZenML, Building Durable AI Agents (Practical AI podcast)
- MCP●●●●●ZenML MCP server, cdtm-mcp
Experience
- 2021–nowCo-founder & CTO, ZenML
I co-founded ZenML and lead its engineering: the open-source framework, the MCP server and Kitaru for durable agents.
- Co-founder & CTO, maiot
Deployed ML in production for predictive-maintenance use cases. ZenML grew out of what we learned there.
- Machine Learning Engineer, Motius
- AI Engineer, Digital Product School
Answered in advance
What do you want to talk about?
Agents in production, evals, and machine learning in production. The more concrete the problem, the better.
What is ZenML?
One AI platform from pipelines to agents. It's open source: github.com/zenml-io/zenml.
Can I pitch you a product or a marketing service?
No. I'm not open to sales or marketing pitches.
Where have you spoken before?
Most recently on Practical AI (Building Durable AI Agents) and the MLOps.community podcast (Agents are Just While Loops), and at MLOps Community London on building a vendor-agnostic AI platform. Before that: an MLOps panel at the Future of Data & AI Conference, Data Science Dojo's LLM Bootcamp, and ZenML's own podcast, Pipeline Conversations.
What did you do before ZenML?
I co-founded maiot, where we put ML into production for predictive maintenance. ZenML came out of that. Before that I was an ML engineer at Motius and an AI engineer at Digital Product School.