mohith
Mohith
ENGINEER · OPEN SOURCE BUILDER

Hey, I’m Mohith.

Building the things behind the screen.

I build infrastructure for AI systems: schedulers, guardrails, and developer tools that decide what actually runs.

GSoC ’26 contributor at RUXAILAB , previously a product developer intern at Koncepts. Currently building Kestrel, a GPU compute control plane.

Experience

A few places I’ve built, learned, and taken responsibility.

THE FOUNDATION

Chinmaya Vishwavidyapeeth

B.Tech, Computer Science

Sep 2023 to May 2027CGPA 9.85 / 10.0

Work

04 PROJECTS

AI infrastructure, developer tools, and a few things I wished existed.

Latest writing

All posts →
8 min read

Reward Hacking: When the Model Learns the Metric Instead of the Task

Notes on reward hacking in reinforcement learning and RLHF: what it is, why RL is so prone to it, Goodhart's Law and its four flavors, how it shows up in LLMs (sycophancy, evaluator hacking, in-context reward hacking), and the mitigations that exist today.

Reinforcement LearningRLHFAI SafetyReward HackingLLMs

Activity

0
Contributions
0
Merged PRs
0
Stars earned
0
Public repos

0 day streak·longest 0 days

Stack

Every day

What I open without thinking about it

PythonTypeScriptFastAPINext.jsPostgreSQLDockerGitClaude Code

Often

Reached for whenever the problem calls for it

ReactKubernetesRedisTailwindSupabaseVuePrometheusVercel

When it fits

Used in anger at least once, happy to again

JavaDjangoSpring BootLangChainOllamaFirebaseMySQL

Learning

  • Kubernetes & Kueuescheduling, quota, and operators from the inside
  • Distributed systemsmetering, admission control, backpressure
  • GPU infrastructureallocation, autoscaling, cost modelling
  • Model evaluationfaithfulness, hallucination, and PII scoring
A line that stays with me