cv

Curriculum Vitae

Sohan Venkatesh

Education

Bachelor of Technology, Computer Science 2022 — 2026
Manipal Institute of Technology, Bengaluru

Experience

Research Fellow Jul 2026 - Present
LASR Labs
Research Intern May — Oct 2025
IIIT Hyderabad
AI Fellow Jul - Sep 2024
Fellowship.ai
Project Intern May - Jul 2024
Infosys

Publications

S. Venkatesh, A. M. Kurapath. "On the Non-Identifiability of Steering Vectors in Large Language Models." Representational Alignment Workshop at ICLR 2026. [Paper]
S. Venkatesh. "Architecture, Not Scale: Circuit Localization in Large Language Models." Mechanistic Interpretability Workshop at ICML 2026. [Paper]
S. Venkatesh. "Negative Before Positive: Asymmetric Valence Processing in Large Language Models." Mechanistic Interpretability Workshop at ICML 2026. [Paper]
S. Venkatesh. "Repeated-Token Counting Reveals a Dissociation Between Representations and Outputs." Preprint. [Paper]

Research

Do LLMs Actually Feel Happy for You? — A mechanistic look at how LLMs process emotional context. [Blog]
Inverse Scaling in CoT faithfulness — Showed unfaithfulness in larger LLMs is a capability artifact after controlling for accuracy. [Code]
Election Safety Benchmark — Pilot benchmark for testing if LLMs give voters accurate election information. [Code]
Prisoner's Dilemma in LLMs — Analyzed three different LLMs with personality variations to study their behavior and cooperation. [Code]

Projects

nanoKimi — A minimzed implementation of Kimi-K2 with custom Muon optimizer and latent attention mechanism. [Code]
anada — A lightweight, terminal-first note-taking tool with Markdown support and bi-directional linking. [PyPI]
QueryMind — An AI agent for natural language to SQL query translation via MCP. [Code]
MCRAG — A multi-code review and generation system. [Code] [Video]

Achievements

  • $700 BlueDot Impact grant for CoT faithfulness research
  • $2,800 Travel Grant for ICLR 2026
  • $1,500 Cohere Labs Catalyst Grant
  • Amazon ML Summer School'26
  • AWS AI&ML Scholarship 2026

Volunteering

Reviewer — COLM 2026, ACL TrustNLP 2026, ICLR Re-Align and CAO Workshops 2026