Backend / AI-Infrastructure Engineer · Pune, India
I build the control plane for AI agents.
What an agent is allowed to do, and whether you can trust what it concluded. Below is a running service graph. Break it, and watch an agent diagnose the incident you just caused — then watch it keep that answer after you take the evidence away.
A live service graph: ingress → api-gateway → checkout-api → postgres/redis, with traffic flowing between replicas. Inject a fault and an agent investigates it.
- waiting for an incident
Not a video and not a mock-up — a service graph simulated in the page, with the fault types and evidence taken from KLRB's scenarios. Open to backend / AI-infrastructure / platform roles.
The system
Three of these are not separate projects. They are one control plane, built in the order I discovered I needed them — and they call each other. Every box below is a repository you can read.
a request from an AI agent
Atlas authorize Scoped, attenuable capability tokens bound to SPIFFE workload identity. Verified fully offline. ★ 40 attenuating — narrower on every hop, never wider AISRE act Detects a live failure, investigates with a three-tool agent, proposes a root cause. new no remediation without human approval structured-JSON root cause KLRB evaluate Did the model read the cluster evidence, or guess confidently without it? ★ 96a conclusion you can check
Adjacent — same question, one layer down
Terraform tells you what drifted. This tells you who changed it, when, and from where — by correlating the plan against CloudTrail.
Atlas asks whether a caller may act. tf.why asks who already did. It does not wire into the pipeline above — it is a standalone CLI, and I would rather say so than draw an arrow that isn't there.
Earlier work
When a test fails, an agent reads the failure, writes a fix, and opens a pull request — inside the same pipeline run that caught it.
Go and Svelte tooling for standing up an application stack without hand-wiring it.
There are ~60 repositories on my GitHub. These six are the ones worth your time; the rest are coursework, forks and experiments, and I would rather point you at four good things than forty.
Background
Education
MIT ADT University, Pune
B.Tech, Electronics & Computer Engineering
Aug 2022 – May 2026 · CGPA 7.71 / 10
Certifications
- AWS Certified Solutions Architect – Associate
- AWS Certified AI Practitioner
- AWS Certified Cloud Practitioner
- GitHub Copilot Certified
- Redis Certified Developer (Python)
Skills
Hiring for backend, platform, or agent infrastructure?
I am looking for a first role where the hard part is correctness under adversarial conditions — authorization, evaluation, incident response. Happy to walk you through the 12 deployment defects the Atlas audit found, or why KLRB's ablation is the only honest way I know to score an AI incident responder.