Co-Founder & CTO, Hezarfen LLC · IEEE Senior Member (CIS) · AAAI Member · Seattle, WA
cagritemel.com · ORCID · LinkedIn · Medium · Contact
Some of the strongest methods in machine learning never leave the paper they were published in. No implementation, no documentation, no pip install. I write the missing half, and I make the result explainable enough to use where being wrong is expensive: a robot that moves, a model a clinician reads, a decision a regulator will ask about.
| neural-trees | Soft decision trees, hierarchical mixture of experts and the 5×2cv F test, behind a scikit-learn API on a PyTorch backend. A gap in a table is not a result until it survives a significance test. | pip install neural-trees |
| reasongate | A self-hostable gate for LLM applications. Answers allow, flag or block with the reason and an audit record attached. Pure Python, zero dependencies, no network calls. |
pip install reasongate |
| ml-playground | sklearn, XGBoost, LightGBM, CatBoost, TabNet and neural-trees compared side by side, with significance testing built in. | |
| ml-academy | An interactive curriculum that teaches machine learning, deep learning and large language models from zero. 123 lessons across 5 tracks, 401 interactive steps, 165 visualisations, in English and Turkish. Runs entirely in the browser: no install, no build step, no server. | mltraining.org |
| AI365 | Daily AI, ML and robotics builds. Open and community driven. |
Verification and auditability for AI systems deployed where the output has consequences.
- TRACE - every autonomous action traced back to sensor evidence through a documented causal chain. IEEE SoutheastCon 2026, pp. 1-6 · doi · arXiv · what it is, in plain terms
- CT-SAFR - safe and interpretable chain-of-thought reasoning for autonomous robots. IEEE CAI 2026, pp. 598-603 · doi · arXiv
- Domain-specific vs general-purpose LLMs in orthodontics - a blinded comparison against GPT-4o, Gemini and Llama. Dentistry Journal 14(4):219, 2026 · doi
- CogniTest - an empirical framework for LLM-assisted test automation. Preprint · doi
Patents - USPTO 63/975,114, safety-constrained chain-of-thought for autonomous systems, sole inventor, pending · TR 2019 22779 B, granted.
Service - Technical program committee: IEEE BigData 2026 (Phoenix, December) · AAAI/ACM AIES 2026 (Malmö). Proposal review: SciPy 2026, 243 proposals · PyData Global 2026, 157 proposals (NumFOCUS). Reviewer: IEEE SMC · CAI · ECCE 2026. Organizer of the special session on trustworthy and explainable AI for telepresence and autonomous robotic systems, IEEE Telepresence 2026. Session chair, IEEE SoutheastCon 2026.
Judging & mentorship - Industry mentor, Penn × Anthropic Sprint Hackathon (220+ participants, 130+ teams) · Judge, Vibe Space @ SF Tech Week · Mentor, Opportunity Hack / ASU · Jury member, ANTSPARK pre-incubation programme.
Full record, talks and CV at cagritemel.com.