About · Research
Research vision.
I am a PhD candidate under the joint supervision of Prof. Elliott Ash and Prof. Mrinmaya Sachan at ETH Zurich, and Prof. Markus Leippold at University of Zurich. I divide my time equally between both institutions.
I study how to make LLMs and agents trustworthy when human instructions are incomplete and success is hard to verify. My research evolves natural-language guidance for more reliable decisions and uses uncertainty to understand when model judgments deserve trust. My long-term goal is to make advances in AI capability translate into better human oversight. I want models to help us recognize where our requirements need clarification and when they lack the evidence to act.
Selected work
Research highlights.
Co-DETECT: Refining natural-language guidance for judgment
Uses annotation uncertainty to surface edge cases, then induces generalizable rules for human review, refining incomplete natural-language guidance for more reliable judgments.
Trace2Skill: Learning natural-language guidance for action
Uses induction across successful and failed executions to improve natural-language guidance for agents, producing reusable skills that transfer across tasks and models without parameter updates.
DIRAS: Calibrating judgments under natural-language guidance
Improves uncertainty calibration through better natural-language guidance, then distills calibrated relevance judgments into smaller models for more efficient deployment.
ReProbe: Using internal uncertainty signals to guide reasoning
Reads internal uncertainty signals from a frozen LLM to estimate reasoning-step credibility, enabling efficient step verification and test-time search without a separate large process reward model.
Projects · Talks · Service
Community collaboration and services.
Collaboration
Apertus: Democratizing Open and Compliant LLMs
Community collaboration on democratizing open and compliant LLMs for global language environments, with contributions to trustworthiness post-training.
Technical report →Collaboration
When AI Benchmarks Plateau
Community collaboration accepted to ICML 2026 on benchmark saturation and how plateauing scores affect evaluation practice.
ICML 2026 →Workshop
ClimateNLP 2024 / 2025 / 2026
Co-organizing the ClimateNLP workshop series: ACL 2024 in Bangkok, ACL 2025 in Vienna, and EMNLP 2026 in Budapest.
Invited Talk
Keynote at the AI4Law Workshop, ICML 2026
Keynote speech at the AI4Law workshop at ICML 2026 on uncertainty-aware LLM reasoning for legal tasks.
Slides →Publications
Selected research
Natural-language guidance and uncertainty for more reliable judgments, agent actions, and human oversight.
ReProbe: Efficient Test-Time Scaling of Multi-Step Reasoning by Probing Internal States of Large Language Models
Reads internal uncertainty signals from a frozen LLM to estimate reasoning-step credibility, enabling efficient step verification and test-time search without a separate large process reward model.
Trace2Skill: Distill Trajectory-Local Lessons into Transferable Agent Skills
Uses induction across successful and failed executions to improve natural-language guidance for agents, producing reusable skills that transfer across tasks and models without parameter updates.
Can Reasoning Help Large Language Models Capture Human Annotator Disagreement?
Tests whether reasoning lets a model recognize when a question is genuinely contested, instead of collapsing real human disagreement into one overconfident answer.
DIRAS: Efficient LLM Annotation of Document Relevance for Retrieval Augmented Generation
Improves uncertainty calibration through better natural-language guidance, then distills calibrated relevance judgments into smaller models for more efficient deployment.
Co-DETECT: Collaborative Discovery of Edge Cases in Text Classification
Uses annotation uncertainty to surface edge cases, then induces generalizable rules for human review, refining incomplete natural-language guidance for more reliable judgments.
Towards Faithful and Robust LLM Specialists for Evidence-Based Question-Answering
Trains open QA specialists to answer strictly from the evidence they are given and resist being misled when it is noisy or missing — keeping every answer traceable to its source.
Research Mentorship & Supervision
Skill-RM: Unifying Heterogeneous Evaluation Criteria via Agent Skill
Research mentorship at Qwen · Tao Chen
GD2PO: Mitigating Multi-Reward Conflicts via Group-Dynamic reward-Decoupled Policy Optimization
Research mentorship at Qwen · Haotian Liu
Understanding Failures in LLM Reasoning by Learning Structured Representations of Chain-of-Thought
Research mentee · Tommaso Felice Banfi
Unlocking LLM Legal Reasoning with IRAC-Constrained Chain-of-Thought
Master student · Adam Rahmoun
Full publication list
Education
Academic path
PhD @ ETH D-GESS
ETH Zürich & University of Zürich
MSc in Data Science & Machine Learning
University College London
BEng in Computer Science
University of Hong Kong