Chat with Docs: Building custom Chatbots using RAG
Nithish Raghunandanan (~nithish9) |
1
Description:
In today’s world, users are increasingly resorting to the use of AI tools (ChatGPT, Copilot, etc) to solve their coding problems. These AI tools using Large Language Models (LLMs) are quite good at coming up with answers for well-established technologies. However, when it comes to new or ever-changing technologies, the LLMs tend to hallucinate. In such cases, it would benefit to have a custom AI bot trained on the technology documentation. In this talk, I cover how to build a chatbot that can answer questions using the existing documentation using the Retrieval Augmented Generation (RAG) technique.
The talk covers how to ingest the documentation into a Vector database. We will answer questions like how to fetch the documentation, how to filter out relevant text on the page, and how to split the long pages into suitable smaller segments for better search results.
After ingesting the documentation, we learn how to build the chatbot using LangChain, an open-source framework for building LLM-powered applications. We learn how to fetch the relevant documents from the vector store based on the users’ chat and generate answers based on the ingested documents. We also handle the challenges in handling the follow-up questions from the user and sources for the answers.
The talk will include practical tips and tricks to help fine-tune chatbots.
We will also discuss how to measure the relevance of the answers and gather feedback from the users to improve the model.
After this talk, the audience will learn how to create a chatbot on their custom data using LangChain.
Prerequisites:
Basic knowledge of Python
Content URLs:
Speaker Info:
Nithish Raghunandanan is a Developer Advocate at Couchbase who loves to build products that solve real-world problems in short spans of time. He has experienced different areas of the industry having worked in diverse companies in Germany and India. Apart from work, he likes to travel and interact and engage with the tech community through Meetups & Hackathons. In his free time, he likes to try stuff out by hacking things together.