Skip to content
View K-kiron's full-sized avatar
:electron:
:electron:
7 followers · 17 following
  • Montreal

Achievements

Achievement: YOLOAchievement: QuickdrawAchievement: Pull Sharkx3

Achievements

Achievement: YOLOAchievement: QuickdrawAchievement: Pull Sharkx3

Highlights

  • Pro

Block or report K-kiron

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
K-kiron/README.md

Open K-kiron's interactive research terminal — explore research, skills, and model pathways.

Open interactive terminal ↗  ·  Try research or skills, then trace a decision.

~ / whoami

I'm Wenhao XU, a Machine Learning PhD student at Université de Montréal & Mila, working on rule alignment for language models.

Do models actually use the rules we give them? I study how explicit rules shape model decisions, with an eye toward AI we can understand and trust.

AI Safety  ·  Alignment  ·  Trustworthy AI  ·  Mechanistic Interpretability

~ / skills

I build open-source agent skills that turn "looks good" into something you can inspect, replay, or challenge.

Skill What it does
RageClick Break a web app like an impatient user. Replay the failure.
SkillClash Find conflicting agent instructions. Trace them to the source.
RepoQuest Turn a real bug into a playable debugging mystery.
PaperCourt Put empirical ML claims on trial against their evidence.
ProveIt Check that a regression test actually catches the bug.

Play RepoQuest →  ·  Watch RageClick →

~ / also-building

TaxAgent — local tax preparation   /   ReviewBudget — pull request verification planning.

Montreal   /   LinkedIn   /   All repositories ↗

Pinned Loading

  1. aims-alignment aims-alignment Public

    Forked from mila-studios/aims-alignment

    Research fork: rule-aligned LLM training and evaluation for human-verifiable compliance assessments.

    Python

  2. TaxAgent TaxAgent Public

    Open-source local tax preparation preview for bounded Quebec and federal salary/student cases.

    Python

  3. ReviewBudget ReviewBudget Public

    Plan pull request verification with named checks, explicit costs, and shared review budgets. Local-first CLI, GitHub Action, and offline HTML reports.

    Python

  4. Intended-Sarcasm-Detection Intended-Sarcasm-Detection Public

    Llama 2 experiments for intended sarcasm detection using data augmentation and prompting strategies.

    Jupyter Notebook

  5. Pretrained-LLM-comparative-study Pretrained-LLM-comparative-study Public

    Comparing Mamba and Pythia across NLP tasks, fine-tuning, and zero/few-shot evaluation.

    Jupyter Notebook

  6. NHL-Analysis NHL-Analysis Public

    NHL data analysis project with a Docker-based application.

    HTML 1