Reza Khan Mohammadi

📘 Vision-language models that know when they're guessing. BICR was accepted to the AACL-IJCNLP 2026 main conference. It flags answers driven by language priors rather than the image, at zero added inference cost.PaperCode

📄 Spotlight at ASTRO 2026. Our work on safely automating oncology chart review with calibrated confidence was selected as a spotlight at the ASTRO 2026 Annual Meeting.

🎓 Teaching students across MSU when to trust AI. My work on what an AI system's confidence can and cannot tell us is featured in MSU's new university-wide AI Literacy course, part of the AI-Ready Spartans initiative.WatchAI-Ready Spartans

🎤 Three abstracts at ASTRO 2026, including an oral in the top 11%. The oral shows how risk-informed training reduces catastrophic errors when language models make oncology decisions. Two more abstracts cover selective automation of chart review and clinical validation of a privacy-preserving model for CTCAE toxicity grading.

🎤 Invited talk at JPMorgan AI Research. I presented CCPS, our method for calibrating LLM confidence, to the JPMorgan AI Research team.

📘 How reliable are confidence estimators for reasoning models? RMCB was accepted to the EACL 2026 main conference (20.1% acceptance rate). It covers 347K+ reasoning traces from six reasoning models across clinical, financial, legal, and mathematical tasks.PaperCodeData

📰 CCPS in the news. Our work on AI reliability was covered by MSU Today, discussed on WJR 760 AM with Ryan Ermanni, and featured in a TV interview on WILX News 10.

💼 Joining Microsoft as an Applied Sciences Intern. I accepted an offer to join Microsoft as an Applied Sciences Intern in Fall 2026.

🏆 Honorable Mention at the MSU and Henry Ford Health Precision Health Symposium. Our abstract on calibrating LLM confidence for precision health received an Honorable Mention.

🏆 Outstanding Paper Award nomination at EMNLP 2025. CCPS was nominated for the Outstanding Paper Award, placing in the top 0.4% of 8,174 submissions.Paper

🏆 NextGen Scholar at IEEE BHI 2025. Selected as an NSF-EMBS-Google Sponsored Young Professional NextGen Scholar for early-career work in biomedical AI and health informatics.

📘 Calibrating LLM confidence without modifying model weights. CCPS was accepted to the EMNLP 2025 main conference. It cuts calibration error by 55% without modifying model weights.PaperCodeData

📘 Hybrid student-teacher LLMs for toxicity extraction. Accepted at IEEE BHI 2025. A large teacher model guides compact LLMs to extract cancer toxicity symptoms from clinical notes, at up to 48x lower cost than GPT-4o.

🏆 Elected a Full Member of Sigma Xi. Recognized by Sigma Xi, The Scientific Research Honor Society, where full membership is by invitation only.

📘 A practical guide to LLMs for radiation oncology. Accepted in BJR|Artificial Intelligence. It covers transformer architectures and large language models, with a checklist for evaluating NLP models before clinical use.Paper

📄 Privacy-preserving CTCAE grading at ASTRO 2025. Poster on grading post-radiotherapy toxicities with an instruction-fine-tuned LLM that keeps patient data private.Abstract

📄 Parameter-efficient toxicity extraction at AAPM 2025. Poster discussion on privacy-preserving, parameter-efficient fine-tuning for extracting post-radiotherapy toxicities.

📘 Does biomedical research predict small-business funding? Published in IEEE Transactions on Computational Social Systems with JPMorgan AI Research. We link 10,873 biomedical topics to SBIR funding and measure how scientific advances lead industry investment.Paper

📄 A local student-teacher LLM for toxicity extraction at ASTRO 2024. Poster on a privacy-preserving student-teacher LLM for extracting toxicities from radiation oncology notes.Abstract

📘 MambaNet in SN Computer Science. A hybrid neural network for predicting NBA playoff outcomes.Paper

🎤 Oral presentation at ICCR 2024. Iterative prompt refinement with teacher-student LLMs for extracting radiation oncology symptoms, without additional training.Preprint

🎤 Tutorial on LLMs for finance at ACM ICAIF'23. I co-presented a one-hour tutorial on Large Language Models for NLP in Finance: Applications and Challenges.Slides

📘 PGST in Natural Language Engineering. A gender-based text style transfer method for Persian, extended and evaluated on English.Paper

🎤 Rapid-fire talk at IEEE BHI 2023. Identifying fetal biological sex from phonocardiogram signals, accepted as a five-minute rapid-fire talk and poster.Watch the talk

📘 Prose2Poem in ACM TALLIP. The first model to translate Persian prose into poetry, released with a new prose-poem dataset.Paper

🏆 Top Cancer Control Abstract Award at the Henry Ford + MSU Cancer Research Symposium. Our abstract won the Top Cancer Control Abstract Award, and I was invited to present the work.Watch the talk

📄 Toxicity symptom extraction at AAPM 2023. Poster on training a transformer-based NLP pipeline to extract toxicity symptoms in prostate cancer patients after radiotherapy.

📘 COPER at ICWR 2021. A semantic search engine for Persian COVID-19 articles, with PerSICK, the first Persian semantic similarity dataset.Paper