muneeb.
Case study · 2026

Pacific Trips Chatbot

Multi-channel AI chatbot for a Pakistani travel agency. Deployed on WhatsApp, Facebook Messenger, Instagram DMs, and the public website — qualifies leads through a conversational sales funnel and captures contact info for the team.

Year
2026
Type
Client work
Timeline
4 channels · AI sales funnel · Lead capture
Status
Live
Web Chat
WhatsApp
Messenger
Captured Leads

The problem

Pacific Trips was losing leads. Visitors browsed the site and left, WhatsApp enquiries piled up unanswered outside business hours, and the team had no way to qualify interest before a human got involved. They needed something that could talk to customers 24/7 across every channel they use.

What I built

I built a multi-channel AI chatbot on FastAPI and LangGraph that acts as a 24/7 sales assistant. It guides visitors through a conversational funnel — from greeting to destination discovery to phone capture — then hands qualified leads to the team with full context.

  • Conversational sales funnel

    5-stage state machine (greeting → discovery → lead capture → qualification → close) that adapts its strategy based on what it already knows about the visitor.

  • LangGraph AI agent

    Groq Llama-3.3-70B with a bound tool for fetching live trip details from Supabase. Sliding window keeps context tight; summarizer compresses older turns.

  • 4-channel delivery

    Identical AI pipeline serves web chat, WhatsApp Business, Facebook Messenger, and Instagram DMs — each with its own webhook, session store, and signature verification.

  • Background lead extraction

    After every turn, a background task runs AI extraction plus regex fallback to pull name, phone, email, destination, and traveller count into Supabase.

  • Rate limiting & security

    Upstash Redis sliding window (40 req/5min per IP or channel ID), HMAC signature verification on all webhooks, prompt injection hardening, PII masking in logs.

  • Session management

    Redis-backed sessions for web chat (cookie-based) and all messaging channels. Recovers existing lead rows by phone on server restart to prevent duplicates.

  • Destination knowledge & weather

    Built-in knowledge base for Pakistani travel destinations plus live weather lookups to give contextual trip recommendations.

The result

The chatbot handles customer conversations 24/7 across all four channels. Leads are captured automatically — the team gets a qualified contact with destination, phone number, and travel details without lifting a finger. WhatsApp messages that used to go unanswered for hours now get an instant, contextual response.

4
channels
5
funnel stages
36
app modules
15
test files
All projects← back to work