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

Achievements

Achievement: Pull Sharkx4Achievement: YOLOAchievement: Quickdraw

Achievements

Achievement: Pull Sharkx4Achievement: YOLOAchievement: Quickdraw

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 ↗  ·  Ask about my work, explore skills, or trace a decision.

~ / whoami

I'm Wenhao XU, an AI/ML 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

~ / now

I turn questions about model behavior into experiments, and recurring development problems into tools. Lately, that means:

  • Verification tools. Inspect agent and automation behavior, replay failures, and check whether a test catches the bug.
  • Human–computer interaction. Prototype ways to switch between tasks and pick up where you left off.
  • Creative coding. Build small 3D browser experiences around exploration and memory.

~ / skills

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

Skill What it does
RepoQuest Investigate a Python bug in an offline case, then reveal the tested fix.
RageClick Replay impatient-user failures in local web apps.
ProveIt Run the same regression test before and after a fix.
SkillClash Find conflicting agent instructions. Trace them to the source.
PaperCourt Put empirical ML claims on trial against their evidence.

Play RepoQuest →  ·  Watch RageClick →

~ / also-building

TaxAgent — local PDF intake and return preparation for supported Quebec/federal salary and student cases.

ReviewBudget — plan pull request checks around an explicit verification budget.

~ / upstream

freeCodeCamp — clearer Python curriculum hints, merged.  ·  pytest — proposed fix for warning-filter precedence across test scopes.

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. contextveil contextveil Public

    Local text de-identification with consistent entities, multiple file formats, and reproducible privacy and utility evaluations.

    Python

  4. RageClick RageClick Public

    Replay impatient-user failures in local web apps. An agent skill with screenshots, recordings, and regression checks.

    JavaScript

  5. SkillClash SkillClash Public

    Explain task-specific conflicts between agent skills and project rules, with exact evidence and reproducible reports.

    Python

  6. PaperCourt PaperCourt Public

    Trace machine-learning paper claims to evidence, repeatable checks, and falsifying experiments.

    Python