WAffy – Your smart WhatsApp agent

WAffy Home page
WAffy Dashboard

Inspiration

Small businesses rely heavily on WhatsApp to communicate with customers - handling orders, queries, complaints, and feedback.
But their inboxes often turn into chaos. Important messages get lost, follow-ups are missed, and most of the work is still manual like copying messages into Excel or CRMs.

We wanted to build a lightweight, intelligent agent that could organize this chaos and give business owners their time back.

What it does

WAffy is a smart multi-agent system that:

Listens to incoming WhatsApp messages using WhatsApp Cloud API
Classifies messages by category (order, inquiry, complaint, etc.) and urgency
Logs messages into an Excel sheet or HubSpot CRM (based on user setup)
Supports multilingual classification - WAffy understands 40+ languages
Displays everything neatly on a React dashboard
Helps small businesses streamline their WhatsApp workflows without needing to manually manage each message

How we built it

Frontend: React.js for the user dashboard
Backend: Python with FastAPI
Async Tasks: Celery for background jobs
Database/Storage: PostgreSQL and Excel export
APIs: WhatsApp Cloud API for message listening, HubSpot CRM API for ticket creation, Perplexity Sonar API for early-stage classification
NLP: rule-based classification logic with multi-language support
Collaboration: GitHub, Perplexity, and Google Meet

Challenges we ran into

Setting up and validating webhooks with WhatsApp Cloud API
Building a modular agent-based architecture in a short time frame
Managing multilingual classification with consistent tagging
Syncing message flow between listening, classification, logging, and visualization
Deciding the right trade-off between using advanced AI tools and fast, rule-based methods for MVP speed

Accomplishments that we're proud of

Built a functional end-to-end multi-agent system in under three weeks
Successfully integrated real-time WhatsApp Cloud API messages
Implemented multilingual message classification
Logged classified messages to both Excel and HubSpot CRM
Created a working dashboard with filters and download/export options
Delivered a collaborative product despite having never worked together before

What we learned

How to break down a real-world problem into modular agents
The importance of prioritizing key features that deliver value quickly
Setting up and managing secure WhatsApp Cloud API webhooks
Using Perplexity Sonar API for message understanding, and knowing when to pivot to simpler methods
Handling CRM integration using HubSpot’s ticket and contact models
Building NLP pipelines that support multiple languages
How AI tools, used intentionally, can reduce manual work and improve accuracy