The Lab

Validated AI products, built in public, and the playbook behind them.

I′m Hemant Kumar Angajala Hemant Kumar Angajala, a full-stack AI engineer. I ship real products solo, test them against reality before calling them done, and write up what actually happened.

PythonFastAPINext.jsTypeScriptLangChainClaudeWhisperOllamaPostgreSQLRedisDockerAWSPythonFastAPINext.jsTypeScriptLangChainClaudeWhisperOllamaPostgreSQLRedisDockerAWS

What′s actually live right now

Not case studies of finished work. These are running, half-built, and updated as they change.

How I build

Idea to shipped, out loud.

Every project on this site follows the same loop. Not because it′s a framework I′m selling — it′s just what stopped things from dying quietly in a private repo.

01

Validate the assumption

Before writing product code, test the core mechanism against real outcomes. OpenBeat's prediction engine was checked against 3 published A/B tests before the app existed.

02

Build the smallest version

One content type, one persona pipeline, one path that works end to end. Everything else waits until the core claim holds up.

03

Ship uncomfortably early

IRA runs from a dev checkout, not a packaged build. It's already in daily use anyway.

04

Share what broke

The barge-in echo bug, the CSP regression, the layout fixes that took three tries. The failures are the actual lesson.