Back to work

Voice agents

Pramanik

You report the problem. We keep following up.

  • An autonomous civic accountability hackathon prototype.
  • Three parallel agents and a coordinator plan actions from public complaints.
  • Multilingual voice calls, persistent case context, and database-backed follow-ups.
  • 3rd place at the OpenAI DevDay Community Hackathon, Pune.
Pramanik · project demoWatch on YouTube

One issue, one shared case

  • Group related complaints, resident reports, and photos.
  • Keep conversations, commitments, and next steps in one case history.
Pramanik map dashboard showing civic issues and geographic clusters
The map brings reported issues and their locations together.
Pramanik issues dashboard with case cards and status filters
The issues dashboard groups cases by progress and verification status.

Follow-ups that keep their context

  • Call through Bolna and OpenAI Realtime with tools, transcripts, and recordings.
  • Persist case context in SQLite and Drizzle.
  • Use retries, exclusive job leases, duplicate-dispatch safeguards, and deadline reminders.
Detailed Pramanik civic issue view with report, timeline, and follow-up action
A case view keeps the report and follow-up actions beside the map.

A clear entry point

  • Explain the reporting and follow-up workflow on the landing page.
  • Show the agent and voice stack behind the prototype.
Pramanik landing page introducing the autonomous accountability office
The public entry point: report an issue and follow its progress.
Pramanik technology section describing its agents and voice services
The stack behind case planning, voice calls, and follow-ups.

Built with

  • TypeScript
  • Next.js
  • OpenAI Agents SDK
  • OpenAI Realtime
  • Bolna
  • SQLite
  • Drizzle
  • Zod
  • MapLibre
next projectSocratica