Applied AI for microfinance · Grade 11
Loan-recovery voicebot
An AI voice agent that calls borrowers in Hindi-English, verifies them, explains the loan and logs the outcome, built after watching a microfinance team do it by hand.
- of concurrent calls possible on cloud deployment
- Hundreds
- of concurrent calls possible on cloud deployment
- of outcomes logged for compliance
- 100%
- of outcomes logged for compliance
- manager who said yes
- 1
- manager who said yes
01 The person
Why it exists.
At Bargach Finance I saw how much time and budget went into manual recovery calls. I proposed an automated agent to my manager, then joined Ardent Technologies as an intern to build it. It dials, confirms identity, shares loan details, collects the reason for a missed EMI by keypad, offers a reminder or settlement, and writes everything to a log with the call ID.
02 How it works
The mechanism.
- 01Twilio Voice for outbound calls
- 02Whisper (base) for speech-to-text, run locally for latency and accent robustness
- 03Amazon Polly Aditi for natural Hinglish speech
- 04OpenChat local LLM for a strict, compliance-safe dialogue
- 05CSV logging with timestamp and call SID; ffmpeg, dotenv, ngrok for testing
Stack
- Flask
- Twilio
- Whisper
- Amazon Polly
- OpenChat