Skip to main content

Code gotcheap.Judgmentdidn't.

Senior software engineer, ten years in distributed systems. AI agents now write most of the code. I make sure what ships is production grade: I decide what gets built, set the boundaries, and build the guardrails that prove it still works.

See the proofGet in touch

What changed

  1. Typing is not the bottleneck

    An agent can produce a working feature in minutes. It cannot decide whether the feature should exist, where its boundaries sit, or what it must never touch.

  2. Green is not evidence

    An agent under pressure will skip a test or loosen a CI gate, and the build still passes. Review now means checking that the safety net was not quietly cut.

  3. Context is infrastructure

    Every session starts cold. The repository has to explain itself: who may change what, and what was learned last time. That context is code. It is versioned and tested.

years in distributed systems
10
merged pull requests on Staffezee, built agent-first
520+
auditors on a platform I led
5,000+
users on a gig platform I built backend services for
500K+

Proof of work

Open source · MIT

sisu

An agent-governance harness generator. It installs roles, change-control tiers, quality ratchets, hooks, a knowledge base and an audit ledger into a project, so an agent working there can answer one question from the repo alone.

What am I allowed to change here, and how will I know I didn't break it?

View sisu on GitHub

A ratchet only moves one way. Test count, coverage and lint debt hold their ground.

Enforcement honesty

Every guardrail reports which actor it actually stops. A local pre-push hook stops an honest agent, not one that runs git push --no-verify. Branch protection stops a contributor, not a repo admin. The report says so.

No credit for measuring less

A baseline is bound to the exact command, config and scope that produced it. Narrowing what gets measured does not count as fixing what is broken.

Built agent-first

Staffezee

Bench sales management for IT staffing agencies, from submissions to invoices. Built with tiered autonomy: the riskiest changes need human approval, and in the lower tiers agents implement bug fixes and features end to end, with one human review before merge. A reporting agent detects regressions and opens GitHub issues with diagnostics, and fixer agents pick them up and fix them. The governance layer was worth extracting, and became sisu.

Visit staffezee.com
agent definitions
18
agent skills
17

Experience

  1. 2023 - Now

    Independent Consultant/Contractor

    Senior Software Engineer

    Senior contract engineering: distributed systems, full-stack delivery and agent-first workflows.

  2. 2018 - 2023

    Office of the CAG, Government of India

    Senior Software Engineer

    Led the digital transformation of India's Supreme Audit Institution, the counterpart of the US GAO: a Node.js, Express and React platform built from scratch, used by 5,000+ auditors across 1M+ documents. Kafka event sourcing for append-only audit trails and data dashboards in D3, Plotly.js and Python with Plotly Dash. Deployed with Docker and Nginx to government cloud and on-prem servers. Mentored a team of eight.

  3. 2016 - 2018

    Awign Enterprises

    Software Engineer

    Backend services for a gig platform that grew from 10K to 500K+ users: Node.js and Express REST APIs, PostgreSQL schema design, Razorpay and Twilio integrations with retries and circuit breakers, and React and Redux admin dashboards.

  • AWS Certified Developer Associate, 2024
  • PG Certificate in Blockchain Technology, IIIT Bengaluru, 2021
  • BE, Osmania University, 2016

Stack

Languages
TypeScript, JavaScript, Python, Java, Go, SQL, Bash, Solidity
Backend
Node.js (Express, Fastify, NestJS), Spring Boot, Django, Flask, FastAPI; REST, GraphQL, gRPC, WebSockets
Frontend
React, Next.js, Angular, Redux, Tailwind
Data and messaging
PostgreSQL, MySQL, MongoDB, Redis, Elasticsearch, DynamoDB, Kafka, RabbitMQ, SQS, SNS
Cloud and delivery
AWS (Lambda, EC2, EKS, RDS, S3, API Gateway, Step Functions, CloudWatch), Docker, Kubernetes, Helm, Terraform, GitHub Actions
Testing and observability
Jest, Cypress, Playwright, contract testing, OpenTelemetry, Prometheus, Grafana
AI and LLM tooling
Claude Code, Cursor, MCP, multi-agent orchestration, subagents, context engineering (CLAUDE.md, AGENTS.md, skills), RAG, evals

Why I was ready

  • 2016 - Now

    A decade of production systems

    Idempotency, bounded contexts, contracts between services, failure isolation. What keeps microservices honest also keeps a team of agents from stepping on each other.

  • 2018 - 2023

    Five years building for auditors

    The whole domain asks one question: can you prove what happened? Agentic development asks it every day.

  • Late 2022 - Now

    Early to each shift

    I used ChatGPT from its release and moved to agents as they matured. Directing them is delegation at a higher clock speed: clear briefs, narrow scope, and never trusting a summary over a diff.

How I got here

Neural nets as an intern, then JavaScript by hand. The tools kept changing. The production bar did not.

Read the story

Run some code.

A sandboxed JavaScript playground. Your code runs in an isolated iframe that cannot read or change the page around it.

Open the playground