Jiayao (Claris) Gu

PhD Candidate in AI Research

I am a Ph.D. candidate at Mila – Québec AI Institute and McGill University, specializing in Generative AI, Large Language Models and their applications on Science. Representative contributions include (1) a first-author paper introducing a dual-transformer-encoder framework accepted at ISMB (16% acceptance rate); (2) a co-first-authored paper on model-based optimization with Yoshua Bengio, awarded the Survey Certificate at TMLR 2026; and (3) a co-authored ICML 2026 spotlight paper (top 2.2%). My research excellence is recognized through the DNA2RNA Scholarship ($80,000), the FRQ-NT Scholarship ($26,666) for genome-based therapies, the Tinker Research Grant ($5,000), and Cohere Labs' Catalyst Grants. Previously, I received my B.Sc. in Computer Science and Biology from McGill University, graduating on the Dean's Honour List (top 10%). Beyond academia, I have conducted LLM research at Noah's Ark Lab, and served as an intern supervisor at Tiptree Systems. I also received the Rising Star Award at the McMedHacks Hackathon for a 3D medical image segmentation challenge.

Email LinkedIn Google Scholar Download CV (PDF)

Education

Mila - Quebec AI Institute / McGill University Sep 2024 – Present

Ph.D. Computer Science (GPA: 3.86/4.0)

Supervisor: Prof. Yue Li (homepage)

DNA2RNA Scholar Award ($27,000) — top recognition for genomic-based therapeutics research

McGill University Sep 2022 – Sep 2024

M.Sc. Bioinformatics

FRQNT Research Scholarship ($26,666) — recognizing top Quebec graduate students

Graduate Excellence Award ($33,000)

McGill University Sep 2018 – Apr 2022

B.Sc. Computer Science and Biology

Dean's Honour List (Top 10% of faculty)

Research Interests

Generative AI

Agentic Skills Optimization

AI for Science

Model-Based Optimization

Scientific Agents

Deep Learning

Experience

Noah Ark Lab Oct 2025 – Present

AI Researcher · Montreal, Quebec

• Built large-scale benchmarks for test-time skill adaptation.

• Helped build a 512-task benchmark by injecting hidden task rules into six existing large-scale datasets.

• Identified limitations of existing memory/skill adaptation methods. Through literature review and extensive reproduction and failure analysis of 9+ recent baselines from top ML conferences, identified limitations in how existing methods store, retrieve, update, and transfer past experience, leading to novel memory structure design that improves agents' test-time adaptation.

• Contributed to the design and validation of our memory/skill adaptation method. Based on the baseline analysis, contributed to the core design of our method for jointly improving agentic skill storage, effective retrieval, and transfer adaptation. Made core contributions to the paper.

• Translate the agent-skill research framework into the test-time task-conditioned pipeline and support its implementation and evaluation.

Mila - Quebec AI Institute Jan 2024 – Present

Graduate Researcher · Montreal, Quebec

• Leading research and development on biological language models

• Leading research in biological sequence and structure prediction and generation

• Leading scientific agent development