Skip to content
View MukundShankar-dev's full-sized avatar
4 followers · 2 following

Achievements

Achievement: Pull SharkAchievement: Quickdraw

Achievements

Achievement: Pull SharkAchievement: Quickdraw

Block or report MukundShankar-dev

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
MukundShankar-dev/README.md

Mukund Shankar

M.S. Computer Science @ University of Maryland, graduating December 2026.

I work on language models, ML systems, and products. Recently:

  • building and operating Full Time Brief
  • researching long-form language generation and evaluation
  • exploring tool use, model confidence, recurrent reasoning, and robot learning

Selected work

  • Diagnosing Tool-Call Decision Making in Small Language Models
    SFT, DPO, CAI-style supervision, and representation probing for tool-use decisions.
    Code · Report

  • Confidence Probes for Language Models
    Predicting LLM correctness from hidden states and uncertainty signals across 424K examples and multiple model families.
    Code · Report

  • DexVision
    Exploring reusable manipulation skills with simulated dexterous hands and vision-based robot learning.
    Code

  • LoopFormer
    Early-stage experiments on recurrent transformer blocks and depth generalization.
    Code

Portfolio · LinkedIn

Pinned Loading

  1. Confidence-Probes-For-LMs Confidence-Probes-For-LMs Public

    Predicting LLM answer correctness using hidden-state and uncertainty probes across 424K examples and five QA/reasoning benchmarks.

    Python

  2. DexVision DexVision Public

    Vision-based robot learning for reusable dexterous manipulation skills in MuJoCo.

    Python

  3. loopformer loopformer Public

    Studying recurrent transformer blocks for iterative reasoning, error correction, and depth generalization.

    Python

  4. anirudhporuri/Tool-Call-Decision-Making anirudhporuri/Tool-Call-Decision-Making Public

    Experiments on improving how model performs on tool call decisions

    Python