I’m a PhD physicist turned AI-safety researcher and ML builder who enjoys turning mathematical ideas into useful, testable, and well-documented software.
My work sits at the intersection of AI safety, machine learning, scientific computing, document intelligence, retrieval-augmented generation, and model evaluation. I like building systems that are not only accurate, but also interpretable, reproducible, and useful in real-world workflows.
I build intelligent systems with math in the engine room and usability on the dashboard.
- AI-safety experiments probing chain-of-thought faithfulness and reasoning-model behavior
- AI and machine learning systems for classification, prediction, and evaluation
- RAG and document intelligence tools for extracting, searching, and reasoning over complex documents
- Model-evaluation frameworks that go beyond single-number metrics
- Scientific computing projects inspired by physics, optimization, and numerical modeling
- Portfolio-ready software with clean structure, documentation, and practical demos
A pre-registered causal study of chain-of-thought faithfulness in a reasoning model: prefix interventions and 500 resampled continuations test whether visible reasoning reveals when the final answer becomes determined.
A document-intelligence and RAG system for pharmaceutical-style documents, combining extraction, retrieval, OCR-aware workflows, and chatbot-style interaction.
A classifier-evaluation framework that studies ambiguity across decision thresholds instead of relying on one fixed threshold.
A research-style benchmark exploring convergence acceleration and finite-horizon learning-curve prediction for gradient boosting models.
I am currently focused on:
- Empirical AI safety: chain-of-thought faithfulness and reasoning-model evaluation
- Retrieval-augmented generation
- Document AI and OCR robustness
- Applied machine learning systems
- AI agents and tool-using workflows
- Model reliability, evaluation, and uncertainty
- Turning research ideas into clean, reusable software
Before moving deeply into AI and machine learning, I worked in theoretical and computational physics, including quantum materials, molecular transport, open quantum systems, and numerical modeling.
That background shapes how I approach AI: I care about mechanisms, assumptions, failure modes, and whether a model actually behaves the way we think it does. That last question is now the center of my AI-safety research.
- Website: https://drkianmaleki.com
- LinkedIn: https://www.linkedin.com/in/kian-maleki-phd
- ORCID: https://orcid.org/0000-0002-0317-8766