WandAI - Multi Agent, Multi Tool Assistant

Overview

I worked with wand.ai first as a freelancer and then later as a full time employee. My main goal at Wand was to do full stack implementation of chat agent Wandy.

Technologies Used

• Firebase - Cloud functions and Firestore Database.
• Kubernetes - Kubernetes clusters to host python REPL runtime execution environments, the containers had to be kept on a queue system so that they are immediately available.
• Azure GPT - Multiple Azure GPT were deployed to various data centers across the world, and the cloud functions used the Azure GPT in round robin fashion in order to not hit the API limits.
• Vercel / NextJS - The react frontend was built using NextJS and was deployed to Vercel platform.

Features

• Multi Tool Usage - The Wandy chat agent is highly capable with tools ranging from Email and Calendar integration via Gmail and Outlook, python code execution with ability to generate various artifacts like images, PDFs.
• Long Term Memory - The chat agent was able to keep a very long term memory of the chats, and had ability to summarize and keep the old information compact, in order to keep the chat context clean.
• Working with Files - Users can upload any type of file and Wandy can extract knowledge from the files and then can use those knowledge to drive insights or answer user's questions, while providing accurate quotations.

Challenges and Solutions

The main challenge was to create the chat system, which is highly event driven, scalable and maintainable.

Impact

Wandy chatbot is the main product by WandAI. Wandy has demonstrated excellent capabilities and allowed WandAI to sign development agreements with large enterprises and onboard various paid users and companies.