WaBot.ai: An AI-Powered SaaS Platform for Customer Engagement and Business Automation

Abstract

Small businesses in India run much of their customer contact through WhatsApp, yet many of them still answer every message by hand. Enquiries arrive after hours, go unanswered, or sink in a crowded chat list, and what the business learns about each customer lives nowhere except the owner’s phone. This paper describes WaBot.ai, a multi-tenant software-as-a-service platform that connects a business’s WhatsApp number to a large language model (LLM), a business-specific knowledge base, a lightweight CRM, and an appointment system. Messages arrive through the WhatsApp Cloud API, are combined with organization context, and are passed to an LLM through the OpenRouter gateway. A distinguishing component is an action parser that separates ordinary conversational replies from structured business actions, such as booking, cancelling, or rescheduling an appointment, and executes the latter against Cloud Firestore. The parser also stops malformed model output from being forwarded to customers. We report a working prototype built with React, TypeScript, Express.js, and Firebase. Evaluation to date consists of functional and integration testing of the main workflows. Formal accuracy, latency, and user-study measurements have not been carried out and are therefore not reported.

Introduction

WaBot.ai is an AI-powered Software-as-a-Service (SaaS) platform designed to help Indian micro, small, and medium enterprises (MSMEs) manage customer communication through WhatsApp. It addresses common business challenges such as delayed replies, missed enquiries, repetitive questions, manual appointment scheduling, and poor customer information management.

The platform integrates the WhatsApp Cloud API, Large Language Models (LLMs), a business-specific knowledge base, a Customer Relationship Management (CRM) system, and an appointment management module. The AI generates responses based on business information and conversation context, while a structured action-parsing layer allows the backend to execute tasks such as booking, cancelling, and rescheduling appointments. This ensures that customer requests can lead to actual business operations rather than just text responses.

WaBot.ai uses a multi-tenant architecture that allows multiple businesses to operate independently on the same platform. Its technology stack includes React and TypeScript for the frontend, Node.js and Express.js for the backend, Firebase Cloud Firestore for data storage, Firebase Authentication for user authentication, and OpenRouter for accessing language models. A central dashboard supports customer records, conversations, bookings, and coordination between AI and human staff.

The system primarily targets clinics, salons, coaching centres, and restaurants, although its architecture can support other service-based businesses. Its main objective is to automate repetitive customer support tasks, improve response times, organize customer data, and streamline appointment management.

Conclusion

We described WaBot.ai, a WhatsApp-based platform that pairs an LLM with a knowledge base, a CRM, and appointment workflows for Indian MSMEs. The prototype receives customer messages, replies through the WhatsApp Cloud API, and carries out booking operations through a parser that turns model output into backend actions. Functional testing of these workflows was successful. We have no measurements yet of accuracy, speed, or user satisfaction, and we have chosen not to report any. Producing those measurements and finishing the real-time dashboard are the next steps. If the results hold up, the central idea of the work, letting the model interpret and letting ordinary code act, looks like a practical route to automating small-business customer contact.

References

[1] A. Vaswani et al., “Attention is all you need,” in Advances in Neural Information Processing Systems 30, 2017, pp. 5998–6008. [2] T. B. Brown et al., “Language models are few-shot learners,” in Advances in Neural Information Processing Systems 33, 2020, pp. 1877–1901. [3] P. Lewis et al., “Retrieval-augmented generation for knowledge-intensive NLP tasks,” in Advances in Neural Information Processing Systems 33, 2020, pp. 9459–9474. [4] S. Yao et al., “ReAct: Synergizing reasoning and acting in language models,” in Proc. Int. Conf. Learning Representations (ICLR), 2023. [5] T. Schick et al., “Toolformer: Language models can teach themselves to use tools,” in Advances in Neural Information Processing Systems 36, 2023. [6] Google, Firebase Authentication. [Online]. Available: https://firebase.google.com/docs/auth [7] Google, Cloud Firestore. [Online]. Available: https://firebase.google.com/docs/firestore [8] Google, Authenticate Using Google Sign-In with JavaScript. [Online]. Available: https://firebase.google.com/docs/auth/web/google-signin [9] Google, Get Realtime Updates with Cloud Firestore. [Online]. Available: https://firebase.google.com/docs/firestore/query-data/listen [10] Meta, WhatsApp Business Platform Documentation. [Online]. Available: https://developers.facebook.com/docs/whatsapp/ [11] OpenJS Foundation, Express.js: Fast, Unopinionated, Minimalist Web Framework for Node.js. [Online]. Available: https://expressjs.com/ [12] Google Cloud, Quickstart: Build and Deploy a Node.js Web App to Cloud Run. [Online]. Available: https://docs.cloud.google.com/run/docs/quickstarts/build-and-deploy/deploy-nodejs-service [13] Google Cloud, The Node.js Runtime in Cloud Run. [Online]. Available: https://docs.cloud.google.com/run/docs/runtimes/nodejs

Copyright

Copyright © 2026 Manthan Pande, Yasir Sheikh, Bhupendra Bhujade, Chaitanya Nile, Harshit Jangra, Harshal Shahare, Shyam Naik. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.