~/arjunsherugar

LLM + telephony

InfoBridge AI

Describe a customer in plain English and it finds the details, then reads them down a phone line.

How it works

A front-desk assistant for the details people keep asking for. Type or paste a sentence and Gemini pulls out the structured pieces — names, policy numbers, phone numbers, emails and company — instead of making someone search for them. From there the app can place a Twilio voice call and read the information back over the phone.

Engineering detail

  • ▸Natural-language extraction with Google Gemini: unstructured text in, five kinds of structured field out — names, policy numbers, phone numbers, email addresses and company details.
  • ▸Twilio voice integration that places an outbound call and speaks the request through generated TwiML, with call status and call logs tracked per session.
  • ▸bcrypt-hashed authentication with register, login, and a password-reset flow. No password is stored in the clear.
  • ▸Session-scoped storage so extracted details stay grouped by conversation, with a browser UI for search, results and call management.

What it doesn't do

The phone call plays a spoken message and hangs up — it does not connect to a live agent or hold a conversation. It is an early prototype with several parallel backend drafts and no automated tests.