AI and Full Stack Engineer

Ashutosh Jha

I build agent systems and the infrastructure they run on.

About

Full stack engineer with 3+ years building web, mobile, and AI powered products. Polished, fast interfaces on top of backends that stay reliable under load.

I take products from concept to production: architecture and data models, UI and analytics, deployment and everything after. Day to day that is React and Next.js up front, Node and Python behind it, and AI features wired in where they earn their place.

Where I've worked

What I'm doing now

Experience

Insitu

Software Engineer

February 2026 - Present

Bengaluru, Karnataka, India (Remote)

South Park Commons

Software Engineer

June 2025 - January 2026

Bengaluru, Karnataka, India

Transak

Developer Relations Engineer

January 2024 - June 2025

Bengaluru, Karnataka, India

CubeAPM

Software Engineer

December 2022 - December 2023

Delhi, India (Remote)

Internships and early startup work live on the experience page.

Projects

Stub

A drop-in budget gate for AI agents that spend real money.

typescriptawsaurora-dsql

NeverLose

An AI sales agent that works on hesitant buyers in real time.

next.jstypescriptfastapi

FinQuery

An OpenEnv-compatible reinforcement learning environment for financial reasoning.

pythonfastapireinforcement-learning

Evolv

Prompt evolution as genetic sequences.

pythonnext.jstypescript

WinterArc

An offline-first habit tracker for a fixed run of days.

typescriptreactvite

StudioCast

The quality layer for AI audio drama.

pythonllmsaudio

Experiments, packages, and the archive live on the full project index.

Stack

Most weeks I am in JavaScript and TypeScript up front, Node and Python behind it, with agent tooling on top and Postgres underneath.

JavaScriptTypeScriptReactNext.jsReact NativeNode.jsNestJSPythonFastAPIGoPostgreSQLMongoDBRedisLangChainRAGMCPDockerKubernetesAWSGCP

Editors, services, and the rest of the kit live on the uses page.

GitHub activity

Writing

All posts

EvalsLatest

Evals are the product

Ship an LLM feature and the eval suite becomes the spec, whether you meant it to or not. Write it like one from the start.

Agent identity

Giving agents an identity you can verify

The moment an agent takes a real action, someone has to say who asked for it and whether they were allowed. Signed envelopes checked against policy answer both.

Caching

Caching in front of an agent fleet

One cache layer does not survive LLM traffic. Three tiers and an in-flight coalescer held costs and latency down, and the tracing mattered more than any of them.