BACKEND · APPLIED AI · AGENTIC SYSTEMS

Divyang Chauhan

Applied AI engineer building systems you can measure, verify, and ship.

I design agent architectures, evaluation harnesses, and LLM-backed products — bringing the backend, security, and infrastructure depth needed to make AI systems reliable beyond the demo.

View the work →divyang@divyang.devIndia · remote · OSCP
FEATURED_PROJECT[4]

Four projects, built in 2026

CENTERPIECE

Pramana

multi-agent smart-contract auditor · Python

A three-agent auditor built on one thesis: a finding counts as real only when a proof-of-concept exploit actually executes. For every hypothesized vulnerability the system writes a Foundry test that triggers the exploit and runs it — if the test doesn’t pass, the finding doesn’t ship. False-positive rate becomes a mechanical outcome, not a judgment call.

13/14 true positives0 false positivesstructural context isolationoffline CI corpus

ClinchCV

shipped · live

A deployed Next.js app that turns a resume PDF into structured feedback: rubric-based scoring, ATS checks, job-fit analysis against a pasted description, bullet rewrites and cover letters. Every LLM response is schema-validated with retry-on-invalid and model fallback, and the whole thing is covered by 649 tests in CI that run against real infrastructure rather than mocks.

Tarpan

distributed · TS + Python

Reads a death certificate and produces the notification letters an executor has to send — sixteen institution types, from Social Security and the IRS down to streaming subscriptions, each with its own required format and enclosures. Four services across an async queue boundary, with SSNs encrypted at the application layer before they ever reach storage.

Shruti

platform · C# / .NET 8

Windows-native dictation that runs entirely on the machine. A global hotkey captures microphone audio, a bundled whisper.cpp build transcribes it with no network call, and the text lands in whichever window was focused before recording started. That last step is the hard one: inserting text reliably into arbitrary third-party applications, each with its own idea of how input arrives.

Also: Jobsieve (job aggregator, NestJS) · Verdikt (alt. Kleros Court UI) · Mushak (Rust, in winget) · DiffLoom.

STACK & DEPTH

What I actually work in

LANGUAGES

TypeScript · Python · JavaScript · SQL · Solidity · C# · Rust

FRONTEND

React · Next.js · Angular · TypeScript · Tailwind CSS

BACKEND

NestJS · Django / DRF · Express · GraphQL · Pydantic · TypeORM

AI SYSTEMS

Agent orchestration · provider abstraction · forced-tool schemas · output validation · context isolation · eval harnesses · local inference

DATA & INFRA

PostgreSQL · MongoDB · MySQL · AWS · Terraform · CDK · Docker · Kafka · Celery

SECURITY

OSCP · Slither · Foundry · Halmos · Echidna · smart-contract review

DEPTH & RANGE

Deepest: backend system design, API design, database modeling & migrations, async processing, event-driven architectures.

Working breadth: React and Next.js frontends, AWS infrastructure, Docker, Terraform, the Web3 stack, Windows desktop development.

Weakest: CSS and pixel-level frontend — can read, debug, and ship clean UI changes; leans on AI for styling.

Mindset: stacks are tools. The architecture and the problem matter more than the medium.

GET IN TOUCH

Let’s talk about the hard part.

Backend and applied-AI roles. Based in India — working remote. Email is the fastest way to reach me.

divyang@divyang.dev →

© 2026 Divyang Chauhan · built with precision, not buzzwords