Cagri Temel
AI/ML Engineer · Researcher · Co-Founder & CTO, Hezarfen LLC
I build trustworthy, explainable AI: safe Chain-of-Thought reasoning for autonomous robots, and interpretable machine-learning systems for high-stakes domains.
About
I am an AI and machine-learning engineer and researcher, and the Co-Founder and Chief Technology Officer of Hezarfen LLC, where I lead Vardenus, an AI-driven real-estate (PropTech) and LegalTech platform built on artificial intelligence, blockchain, and modern cloud infrastructure. My work sits at the intersection of explainable AI, AI safety, and applied large-language-model systems, with a focus on making advanced models trustworthy: grounded, auditable, and safe enough to deploy in high-stakes settings.
As an IEEE Senior Member and a Senior Member of the IEEE Computational Intelligence Society, my current research centers on safe and interpretable Chain-of-Thought reasoning for autonomous robots. This work has been presented at IEEE venues including the IEEE Conference on Artificial Intelligence (CAI 2026) and the IEEE New Era AI World Leaders Summit. I serve the community as a reviewer and program-committee member for several IEEE and AAAI/ACM venues.
I hold an M.S. in Computer Science from Grand Canyon University (GPA 3.89, Alpha Chi Honor Society) and a B.S. in Electrical and Electronic Engineering from Istanbul Aydın University. I am an inventor on patents in both the United States and Turkey, and I am based in Redmond, Washington.
Teaching is the other half of the work. I built and wrote ML Academy (mltraining.org), a free and open-source curriculum of 123 interactive lessons that teaches machine learning, deep learning, and large language models entirely in the browser, in English and Turkish. It is built on one idea carried over from my research: a course should not just show machine learning, it should make the student prove it.
I am applying to full-time PhD programs in the United States. The question I want to spend those years on is how a safety claim about a reasoning system gets established rather than asserted: which safety properties survive when a model's account of its own reasoning is unfaithful, what continuous auditability costs the system being audited, and how safety results should be reported so a reader can tell a strong one from a weak one. I am looking for a group where verification and evaluation methodology sit close to real autonomous systems, and where a result that fails its own controls is worth publishing. Mine did fail, this year, and writing that up taught me more than the original result would have.
Experience
-
2020–
Co-Founder & Chief Technology Officer
Lead the design and implementation of the PropTech and LegalTech systems behind Vardenus, an AI-driven rental platform connecting landlords, tenants, and contractors: an LLM-driven legal-automation module (R-Law) for landlord-tenant mediation, compliance, and dispute resolution, and security-first MLOps infrastructure. Tokenized fractional ownership and escrow automation on Polygon PoS were designed and are the subject of a U.S. patent application.
-
2025–
Machine Learning Engineer
Ship LLM-powered features end-to-end, from data pipelines to training, evaluation, and inference. Built retrieval-augmented generation with guardrails for higher answer quality and reliability; a model release process with CI/CD, canary and A/B releases, and drift monitoring; and APIs with FastAPI and Docker, with latency and cost optimization.
-
2020–
Software Development Engineer in Test
Designed and automated test suites with Java, Selenium WebDriver, TestNG, JUnit, and Cucumber (BDD / Gherkin); data-driven testing, the Page Object Model pattern, and API testing with Postman and REST Assured.
-
2019–20
Maker Engineer
Led STEM and maker-space initiatives and hands-on curricula, mentoring 100+ students and organizing community showcase events.
-
2018–19
Quality Assurance Engineer
Architected QA and test-automation frameworks (Java, Selenium, TestNG; TDD/BDD) for web-based educational platforms with video streaming and student-tracking systems.
Education
-
2025
M.S., Computer Science
Specialization in advanced machine learning and artificial intelligence. Capstone project: large language models applied to real estate workflows (Integrated Real Estate Ecosystem Platform), advised by Dr. Aiman Darwiche.
-
2018
B.S., Electrical and Electronic Engineering
Graduation design project: High Gain Antennas for CubeSatellite, an S-band microstrip patch antenna on FR-4, modelled and simulated in ANSYS HFSS. Advisor: Asst. Prof. Saeid Karamzadeh.
-
2017
Space Studies Program
Certifications: Machine Learning (MIT) · Machine Learning Professional (IBM) · Deep Learning & Neural Networks with Keras (IBM) · Generative AI with LLMs (AWS / DeepLearning.AI)
Skills
AI & Machine Learning
Deep learning and neural networks (computer vision, NLP, generative AI, LLMs), explainable AI, predictive modeling, robotics AI and Chain-of-Thought reasoning.
Software Engineering & Testing
Python, Java, C/C++; test automation (Selenium, TestNG, JUnit); TDD/BDD; CI/CD (Jenkins, GitHub Actions).
Cloud & MLOps
AWS (EC2, Lambda, S3, RDS); Docker, including native arm64 ROS images; SQL and NoSQL; API design and integration.
Leadership
Cross-functional and global team leadership, strategic technology planning, and STEM program development.
By the Numbers
Honors & Awards
-
2019
Best Space Project, Maker Faire San Francisco
-
2016
Second place internationally, CanSat Competition (AAS / NASA / JPL)
-
2025
Alpha Chi Honor Society
Affiliations & Recognition
IEEE Senior Member· IEEE Computational Intelligence Society· AAAI· International Space University· Grand Canyon University
Upcoming
- Nov 9–13, 2026Organizing the special session on Trustworthy and Explainable AI for Telepresence and Autonomous Robotic Systems at IEEE Telepresence 2026, Bristol, UK.
- Dec 14–17, 2026Serving on the Program Committee of IEEE BigData 2026, Phoenix, AZ.
News
- Aug 2026On the humanoid sprint record set in Beijing, and why the padded barrier at the finish line is the more informative half of the result. Write-up.
- Aug 2026Measured why a terrain classifier could not drive the policy: the signal was in the data but arrived 8.3 seconds into a ten second crossing. Write-up.
- Aug 2026Reviewed talk proposals for PyData Global 2026, the NumFOCUS Python data-science conference, after serving as a proposal reviewer for SciPy 2026 in July.
- Aug 2026Wrote up what five review assignments taught me about results a reader cannot falsify, using my own retracted slip study as the example. Write-up.
- Aug 2026Joined the Program Committee of IEEE BigData 2026, the IEEE International Conference on Big Data, in Phoenix this December.
- Aug 2026Found and fixed a rig fault in my ROS 2 slip study: the robot had been resting on its front edge at 13.3 degrees, which manufactured the effect being measured. Slip was re-established on a controlled ramp, and the earlier post carries a correction. Write-up.
- Aug 2026Released ML Academy at mltraining.org: 123 free interactive lessons on ML, deep learning, and LLMs, open source and in two languages.
- 2026Organizing a special session on Trustworthy & Explainable AI at IEEE Telepresence 2026, Bristol.
- May 2026CT-SAFR presented at the IEEE Conference on Artificial Intelligence (CAI 2026) and published in IEEE Xplore.
- May 2026Taught the hands-on Data & Analytics workshop on explainable neural trees at Washington State University.
- May 2026Joined the Program Committee of AAAI/ACM AIES 2026 (AI, Ethics & Society).
- Apr 2026Paper published in Dentistry Journal (MDPI, Q1, IF 3.1): a blinded comparison of AlimGPT against GPT-4o, Gemini, and Llama.
- Feb 2026TRACE published at IEEE SoutheastCon 2026, where I also chaired the AI and Predictive Modeling session.
- Feb 2026Spoke on “AI That Matters: Trust Over Power” at Louisville AI Week 2026.
- Dec 2025Invited speaker at the IEEE New Era AI World Leaders Summit, Seattle.
Teaching: ML Academy
mltraining.org is a free, open-source curriculum I wrote from scratch: 123 interactive lessons that run entirely in the browser, with no videos, no installation, and no payment. Other courses show machine learning. This one makes you prove it.
Predict, then see
The student commits to an answer before any animation runs, and the hit rate becomes a calibration score that exposes where intuition fails.
Feel the need for the tool
Re-split the data with a new seed, watch the model ranking flip, and only then meet the 5×2cv F-test that settles it.
Run your own answer
Lessons end in executable code. A wrong sign is not rejected, it runs, and the student watches gradient ascent blow the loss up.
Every number verified
123 lessons, 401 steps, 165 visualizations, 387 cited references, and a script that re-derives every figure before release.
Research Areas
Explainable AI (XAI)
Making model decisions inspectable and auditable through traceable reasoning, grounding, and governance for AI used in regulated, high-stakes settings.
Safe Chain-of-Thought Reasoning
Multi-layered verification that treats LLM reasoning as a checkable control artifact, detecting unsafe or hallucinated steps before any action.
Trustworthy Autonomous Systems
Decision frameworks for robots that trace every action back to sensor evidence, built for auditability under the EU AI Act and ISO 13482.
Interpretable ML / Neural Trees
Architectures that combine neural networks with decision-tree transparency for robust, explainable predictions under noise and missing data.