Doruk Ozar
Founder of Diador. Applied AI engineer.
Claude Certified Architect — FoundationsDoruk builds AI systems for regulated industries — healthcare and insurance, where an answer that merely sounds right isn't good enough and every step has to survive a compliance review.
It's an unusual place to learn this craft. Most AI gets built where mistakes are cheap. His has been built where they aren't: multi-agent systems doing document-heavy knowledge work, retrieval across hundreds of thousands of pages of source material, and audit trails written for the people who have to sign off — not for the developers who built them.
He started Diador in 2022, and the premise hasn't changed since. The businesses that would gain the most from this work — accounting firms, law practices, medical offices, service operators — are precisely the ones who can't hire a team to build it. Diador is how they get it anyway, held to the same standard.
What he actually builds.
In plain terms, without the vocabulary that usually surrounds it.
Multi-agent systems
Several AI agents each handling one piece of a larger job and passing work between them — sized to the task, so a simple workflow never gets an over-engineered answer.
Retrieval at scale
Software that searches a company's own documents — hundreds of thousands of them, some running to hundreds of pages — and reasons over what it finds, instead of guessing from memory.
Agent memory
Deciding what a system carries between steps and between sessions, and — just as deliberately — what it is built to forget.
Guided-interview agents
Systems that ask the people who own a process the right questions in the right order, then turn knowledge that lived only in their heads into finished written documents.
Audit trails
Every decision recorded in a form a compliance reviewer can actually read. Built to survive governance review, not to look good in a demo.
Evaluation harnesses
Test suites that gate every change. When a prompt or a model gets swapped, the suite says whether quality moved — nobody has to take it on faith.
“The hardest part of this work was never the model. It's knowing exactly where to stop — what the system decides, and what stays with the person who's accountable.”
Where the three rules came from.
Diador publishes three rules it doesn't break: custom-engineered rather than assembled, humans decide while systems prepare, and bounded systems with visible behavior.
None of those are positioning. They're what regulated deployment teaches you, usually the hard way. When the person reviewing your architecture runs enterprise IT governance, the model figured it out is not an answer. You learn to draw the boundary in writing before you build, to leave the judgement call with the person accountable for it, and to keep a record anyone can read.
Smaller operators almost never get that standard applied to their work. Closing that gap is the whole reason Diador exists.
The path here.
Founder · Diador
Started as a technology consultancy building data pipelines and predictive pricing models for device resale, cutting grading and diagnostic time by 40%. Now an AI-native agency that designs, builds, and operates AI workflows for operations teams.
Applied AI engineering · Enterprise — regulated healthcare & insurance
Lead engineer across concurrent enterprise AI programs, owning architecture end to end: agent memory, orchestrator and sub-agent topologies, retrieval, and the logging and tracing that regulated data demands. Built the agent framework behind a platform where other teams compose their own agents over their own document sets.
Software Engineer · EasyCep
Machine learning models for supply chain cost and smarter inventory placement, plus operational risk detection that surfaced problems before they became disruptions.
Data Engineer · Invent Analytics
Real-time order orchestration and scalable data workflows. One client's lost sales fell from 15% to 3%.
The formal record.
Claude Certified Architect — Foundations
Anthropic
Anthropic's architect certification, examined across agentic architecture and orchestration, tool design and MCP integration, prompt engineering and structured output, and context management and reliability.
Verify on CredlyM.S. Data Science
University of Virginia
Deep Learning · Decoding Large Language Models · Big Data Systems · Applications of Data Science · Linear Models
B.S. Computer Science & Engineering, Economics minor
Bucknell University
Major GPA 3.73/4.0 · Machine Learning · Data Mining · Algorithm Design · Software Engineering
Graduate research — Multimodal RAG Framework
An end-to-end retrieval pipeline spanning text, audio, and video, with a comparative evaluation harness that benchmarked competing architectures and model choices on both accuracy and efficiency.
Engineering Dean's List
A 4.0/4.0 term GPA at Bucknell — earned during a full varsity water polo competition season.
Fifteen years in the water.
Long before any of this, Doruk played water polo — first for the Turkish National Team, then four years of NCAA Division I at Bucknell.
The sport didn't transfer. The habit did: show up, do the unglamorous repetitions nobody watches, and let the results make the argument for you.
- Turkish National Team — Best Attacker Award, Barkalov Tournament
- NCAA Division I varsity water polo, Bucknell University — four years
- ACWPC All-American, Honorable Mention
- ACWPC Scholar-Athlete, Excellent Honors
- MVP of the Year · Rookie of the Year
- MAWPC Championship All-Tournament Team
Talk to Doruk directly.
Every Diador engagement opens with a 30-minute workflow assessment, and he runs them himself. You'll leave with a written view of where your time goes and what's worth automating first — whether or not we end up working together.