About

  • Developed a multi-agent and multimodal chatbot using Python and ReactJs for Clothing retail bussiness for Product recommendation, purchase order and order status. Configured Autogen Agents to generate SQL queries to execute to respond for request from customer.Created complex ERD and Schema on PostgreSQL to let LLM talk to the database efficiently and accurately.

Phase-1: Recommendation Agent

  • Capable of recommending the products based of the users preferences in natural language. It also allows the purchasing of the recommended product if the user wants to buy it and also update in the database with prompt confirmation of order. A user can also see the status of the order based of the orderid
  • Used the GroupChat class in Autogen to effectively orchestrate the conversation between multiple agents by keeping the Human in the loop
  • created Backend APIs using Flask and React for the front end

Demo-1: https://youtu.be/5gD1kuKxsYE

Phase-2: Fraud Detection Agent and Shipping Status Agent

  • Fraud Detection Agent is capable of detecting the fraud based on the OCR image of bill uploaded by the customer. We used gpt-4o-mini for all of the tasks performed by the agents. We also have our agent promptly reply with the status of the fraud by analyzing the image and double checking with values in database
  • Shipping Status Agent is capable of providing the status of the damaged package and defective product by analyzing the image using VisionCapability class in Autogen

Demo: 2: https://youtu.be/kMJhXsEgpmg