Skip to content
Back to the doors

Door 01 / 04

AI & Agents

Agents that reason inside real constraints — grounded retrieval, explicit tools, and a human at the decision point.

How I work

05 steps

  1. 01Frame the failure mode and the human decision it blocks.
  2. 02Ground the model in real records, not guesses.
  3. 03Design tools, guardrails, and approval checkpoints.
  4. 04Test failure cases before a user ever meets them.
  5. 05Instrument, observe, and iterate in production.
RecoverAI payment recovery engine dashboard screenshot

Solo Builder

RecoverAI

AI-powered payment recovery system that analyzes failed payments, recommends safe recovery strategies, applies policy rules, and supports human-in-the-loop approval workflows. Dashboard for cases, human review, execution history, and simulation mode with no real money movement by default.

Human-in-the-loopPolicy engineSimulation mode
PythonFastAPIReact + TypeScriptPostgreSQLDockerRazorpay APIs
Saathi Swasthya voice health triage assistant screenshot

Solo Builder

Saathi Swasthya

Multilingual, voice-first health navigation and triage assistant for Bharat. People in Hindi, Gujarati or English talk through health concerns, get urgency triage, and find real nearby facilities via OpenStreetMap. It is NOT a doctor: it refuses to diagnose, prescribe or rank hospitals, with consent-gated memory, human-escalation tickets and call analytics.

Voice-firstConsent-gated memoryEscalation tickets
PythonLiveKit AgentsDeepgram Nova-3Google Gemini 3.5 Flash LiteMurf Falcon TTSPostgreSQLNext.js 15 / React 19

✓ Murf AI 10 Days of Voice Agents challenge, Health Access track.

Note: Built with the LiveKit Agents starter (Fork) · Local runtime.