I build safety checks and training infrastructure for AI agents.
Member of Technical Staff, Scaler AI Labs · CS undergraduate, BITS Pilani
I'm a member of technical staff at Scaler AI Labs in Bengaluru, where I lead the safety workstream: checks and enforcement that keep frontier models inside their safety guardrails. I also build the infrastructure that trains AI browser agents. I'm doing my CS degree at BITS Pilani. Before this, I was a founding engineer at TrackPrep and a backend engineer at HandaUncle.
Updates
Focus
Checks and enforcement that keep frontier models inside their safety guardrails.
RL environments and automated task generation that exercise real software skills, not toy benchmarks.
Dataset pipelines, validation, and privacy tooling at scale; rubric-based grading for agent work.
Policy gates that check an AI agent's payment before it settles, and guards that keep token integrations safe on Solana.
Work
Member of Technical Staff · Bengaluru
Infrastructure for training AI browser agents. I lead the safety workstream — task design, checks, and enforcement pipelines that verify frontier models follow their safety guardrails.
Also built the automated task-generation pipeline targeting real applications (Gmail, Salesforce, Jira), the dataset export pipeline, and parallel Aho-Corasick PII scrubbing in Rust.
Teaching Assistant
Mentoring 100+ students in data structures & algorithms (Java / JavaScript) and full-stack web development.
Backend Engineer
AI conversational finance. Built multi-model LLM orchestration with retrieval-augmented generation, led the MongoDB → PostgreSQL migration, and built embedding-based analytics pipelines.
Founding Engineer
Built a NEET-preparation platform from day one; owned the backend as it scaled to 85,000+ students. PostgreSQL optimization, Redis caching, and the progressive web app.
Background
B.Sc. Computer Science
Computer science intensive, in parallel · Dean's Merit List 2025
Onchain
Built solo: a trust layer for AI-agent payments on Solana — a policy gate that runs before an agent's payment settles. Three Anchor programs on devnet, formally verified (7 Kani proofs, 662 sub-checks), with a TypeScript SDK and MCP server on npm.
Accept any Token-2022 mint without getting drained: an on-chain guard that refuses hostile mints, a scanner with live mainnet verdicts, and LiteSVM attack tests. Open PRs into Solana AI Kit, alongside its sibling transfer-hook-scaffold.
Builds
An OpenEnv reinforcement-learning environment for SRE incident triage: cascading failures, eight hardening techniques, and binary rubric grading. Finalist at the Meta PyTorch OpenEnv Hackathon.
Natural language → educational animation, built on Manim with LLM-generated scene code. Go + Python + Next.js microservices; ~90% first-pass render success on my test prompt set.
Beyond work
Before any of this, I led my region's basketball team at the national level. The discipline stuck. Fastest way to reach me is email.