← Back to projects
06Hackathon · Payments

AI Teammates

Two AI agents for a payments company: one turns a sales lead into a payment, one fixes a failed payment. Both hand off to a person at set limits.

  1. 01Lead or failed payment
  2. 02AI agent with tool calling
  3. 03Rules: refund cap, deal-size limit
  4. 04Mock tools: payment link, refund, reversal
  5. 05Customer on WhatsApp or SMS
Over the limit, a person gets the case with a full summary. Every action goes to a Postgres audit log.
24.07 bnUPI payments in September 2026 (NPCI)
~0.8%of UPI payments fail for technical reasons
1.33 mnRBI Ombudsman complaints in FY25

Challenge

Sales teams score each lead by hand, and payment links go out slowly. Support teams check each failed payment by hand, then refund, retry or escalate it.

Approach

  • It acts: sends payment links and triggers reversals
  • It hands off: large deals and hard cases go to a person with a case summary
  • It is measured: conversion rate, time to close, cases solved without a person, handle time
  • Business model: used inside the company to lower cost per case, or sold to merchants as a paid add-on, with a small fee per solved case or closed deal

Result

Pitched. The Stage 1 pitch deck was submitted in October 2026.

Event
Paytm Build for India AI Hackathon, Track 3
Team
Tushar Pathak, Manyata Manocha, Sundaram Sharma
Stage
Stage 1 pitch, October 2026