About

Born in Iran in 1980. What follows is a working life assembled mostly out of self-teaching, commerce and long stretches of building things that had to actually run.

Studied computer science at Shahid Shamsipour. He did not complete the degree. The programming came anyway, through practice rather than a certificate: FoxPro first, then Delphi, in the era when a database application was something you wrote, shipped and then supported yourself.

That period set a habit that still shows up in the research: a preference for systems whose behaviour you can inspect, and a suspicion of anything that only works when nobody looks closely.

Work through the Oracle era on database-backed systems. Schemas, migrations, reporting, and the ordinary discipline of keeping records that other people depend on being correct.

Completed military service.

A move into international business: trade across borders, time in the UAE, travel to China, and work around packaging machinery. Sourcing, negotiation, logistics and the unglamorous mechanics of getting goods and money to move between countries.

Running a business teaches a different lesson from writing software. A system that is elegant but unreliable costs somebody real money.

Digital marketing and search work, then the automation that grew out of it. Once you have done the same manual process a few hundred times, you start writing the thing that does it for you.

Full-stack development and automation infrastructure. Frontend and backend, the plumbing between them, and the operational work of keeping deployed things alive.

Hands-on experimentation with generative models: cloud GPU work, running and comparing models, and building real systems on top of them rather than demonstrations.

Building long-running systems on top of language models surfaced a specific failure that did not fit the usual framing. The models were not unintelligent. They handled each task in front of them. What they lost, over long projects, was the thread: the intention behind the work, the decisions already made, the evidence already gathered.

This is the observation the current research is built on. It is a first-person engineering observation, not a published result.

Current work is on the machinery that would let a long-running system keep hold of why it exists: durable state, provenance, and evidence that survives past the end of a conversation. Two strands of that work are described elsewhere on this site, at the level of the problem rather than the implementation.

Alongside it sits a question that is not technical. If automation removes work faster than it creates it, the continuity that matters is not only the machine's.