RAG-based Customer Support Chatbot Development

As part of the 5-day Gen AI Intensive Course with Google, participants had to complete a capstone project. I chose to make a RAG-based Customer Support Chatbot to showcase the skills I learnt**.**

Customer support teams are often overwhelmed with repetitive queries, leading to delayed responses and reduced customer satisfaction. This RAG-based Customer Support Chatbot solves that problem by combining Retrieval-Augmented Generation (RAG) with embedding models, vector search, and a vector database to deliver accurate and relevant answers.

It uses the Customer Support FAQs Dataset, which includes 200 commonly asked questions and their answers. You can also update the list or add your own questions to better match your business.

Behind the scenes, it uses the text-embedding-004 model to understand what the customer is asking. It then searches a ChromaDB database to find the most relevant question-answer pairs. The Gemini 2.0 Flash model uses the user’s query along with the retrieved question-answer pairs to generate the final response.

What sets this solution apart from others that rely on basic keyword matching is its ability to understand the user’s query through embeddings, resulting in more accurate and relevant answers. A RAG-based customer support chatbot like this can help reduce workload, support training for new team members by suggesting how to respond, and ultimately improve the customer experience.

Check out the code here: https://www.kaggle.com/code/shuaibnuruddin/capstone-project