Back to Portfolio
2026
LLMWhatsApp APIPythonEvals

Bizzy — AI WhatsApp Agent

"An LLM-powered WhatsApp agent that onboards and supports thousands of travel agents end to end — reads their KYC documents in chat, answers platform questions from a live knowledge base, and hands off to a human ticket only when it truly can't help. Every release must pass a 170-case AI evaluation gate."

Executive Summary

Agents onboarding to our B2B platform had hundreds of repeat questions and a manual KYC process. I built Bizzy: a WhatsApp agent on our own retention-CRM server that understands text, images and PDFs, collects and checks onboarding documents in the chat, answers from a versioned knowledge base that non-engineers edit from an admin dashboard, and opens a support ticket with full context when a human is needed. Because a small prompt change can quietly bring old bugs back, nothing ships until a 170-case evaluation suite — regressions, prompt-injection and guardrail tests, judged by a second model — passes.

Key Outcomes

  • Thousands of agent conversations handled on WhatsApp without a call centre.
  • Reads PAN / Aadhaar / business documents straight from chat for onboarding.
  • Human handoff with a built-in ticketing desk only when the bot can't resolve.
  • 170-case AI evaluation gate runs before every knowledge-base or model change.

Next Case Study

Festival Demand Prediction Model

A machine-learning model that forecasts seat demand day by day for festival peaks like Diwali and Dussehra — an ensemble of booking-curve and calendar models trained on three years of bookings and a hand-built festival calendar. Within ~7% on Dussehra 2026.