Yuya AsanoPhD candidate in Intelligent Systems, University of PittsburghI build conversational AI that people learn with.
I design, build, and evaluate LLM systems, from agent harnesses for multi-turn dialogue to LLM-as-judge evaluation pipelines and LLM features shipped in industry. At Pitt I built the LLM agent behind Emma, a teachable robot that students teach math to. At SigIQ.ai I built an LLM and retrieval pipeline and shipped a mastery model for a K-12 math tutor. I’m now building an AI learning companion for adults preparing for professional exams.
Finishing my PhD in May 2027 and open to applied scientist, research scientist, and ML engineer roles.
Selected work
- Emma, an LLM teachable agentAn LLM agent that introduces a second solution and helps one student compare the two
- How pairs of students teach a robotLexical alignment and solution-path conflict when two students teach Emma together
- AI learning companionA 3D study partner designed to keep adults studying for professional exams
- Contextual speech recognition correctionFixing mishearings with dialogue context; 3rd place, Alexa Prize TaskBot
- Do LLMs solve math like students?Comparing correct solutions from LLMs and real students, EMNLP 2025
Experience
- SigIQ.ai Machine learning internBuilt a two-stage LLM and retrieval pipeline that labels math lessons with the Common Core standards they cover (+.19–.33 macro F1 over the existing baseline), and shipped a Beta-Bernoulli mastery model as a teacher-facing feature.May – Aug 2026
- Jackson State University NLP research internProposed retrieval-augmented generation that tailors grading rubrics to students’ oral and written answers, and prototyped grading APIs on AWS with LangChain and Docker.May – Aug 2025
- Educational Testing Service (ETS) Ida Lawrence Research InternAutomated scoring of how well teachers lead classroom discussions with GPT-4o. Better task formulation raised correlation with human raters by .12–.28.Jun – Aug 2024
- Alexa Prize TaskBot Challenge 2 Team ISABELContext-aware speech recognition correction that tripled correction recall. Third place.2023
- University of Pittsburgh Graduate research assistantBuilt the LLM agent harness behind Emma and study infrastructure for lab and classroom studies. Advised by Diane Litman.2022 – now
- Carnegie Mellon University Research internClustered student code to deliver programming hints, and prototyped an instructor dashboard that summarizes student reflections.2020 – 2021
- University of Toronto Undergraduate research assistantBuilt a reinforcement learning system that personalized feedback on online learning platforms, and ran field experiments in a live course.2019 – 2021
Recent first-author papers
- EMNLP 2025
- Exploring task formulation strategies to evaluate the coherence of classroom discussions with GPT-4oBEA 2025
- AIED 2025 Code
- COLING 2025, Industry Track
- EDM 2024 arXiv
Education
- University of Pittsburgh PhD, Intelligent SystemsAdvised by Diane Litman. GPA 3.85.2021 – 2027
- University of Toronto Honours BSc, Computer ScienceAI focus. GPA 3.97.2018 – 2021