Interviu
Real-time voice interviews on Gemini Live over bidirectional WebSockets, with credit billing and distributed workers.
TypeScript
React
Node.js
PostgreSQL
Redis
About me
I am an AI engineer focused on building useful agentic systems and live multimodal applications.
Much of my recent work has been focused on making LMs useful. Using LangGraph in Autage I built a multistep incident response agent for Kubernetes, ensuring tool boundaries are clearly enforced to be read-only so prompt injection or hallucination cannot change the state of clusters. In Interviu I removed a transcription and speech step to pass raw PCM audio into the Gemini Live API.
Coming from a background of backend engineering and distributed systems, I often focus on how to handle things outside of the model such as token budgeting, stream failures and state consistency. I enjoy building reliable, safe AI applications.
Selected work
Projects
Autage
A four-stage LangGraph agent that triages P1/P2 incidents, runs read-only cluster diagnostics, and streams its trace live.
Python
FastAPI
LangGraph
Kubernetes
PingDeck
A Go webhook proxy that makes PostgreSQL the queue: 1,000 events/sec, exactly-once claiming, and a live SSE replay deck.
Go
Docker
Skills
The stack and the tools I reach for most days.
Contact
LinkedIn
GitHub© 2026 Vaibhav Kolekar · Built with Next.js