Personal Data
Chatbot
Customized Chatbot | 2024
Client: Various
Skills Used: Python, LlamaIndex, LangChain, Streamlit
The Problem
The request I get most from clients is creating a chatbot based on their own data. The data sources can range from a single PDF file to a data warehouse of documents, spreadsheets, and SQL databases. These documents cannot simply be put into ChatGPT for several reasons:
-
ChatGPT does not provide enough accuracy when analyzing several documents at a time.
-
The data/documents are too large for ChatGPT.
-
The data is sensitive/proprietary and the chatbot needs to be hosted on internal infrastructure.
-
The chatbot needs to use current information/needs to be continuously updated
-
The outputs of the chatbot need to be fed into another software.
-
Other LLMs may perform better than ChatGPT.
-
The chatbot needs to be hosted on an external website
The Solution
The solution to these issues is creating customized chatbots trained on private data using Retrieval-Augmented Generation.
Retrieval-Augmented Generation (RAG) chatbots combine the ability to generate natural language responses with accurate information retrieval from large document collections. They analyze your question, search for relevant data, and generate precise, up-to-date answers. This approach ensures accurate, current, and specialized responses without compromising data security and privacy.
The customized chatbots can be hosted on internal infrastructure, preventing sensitive information being sent externally via API. The outputs of these chatbots can be fed into other internal/external software, handle large amounts of data of various types, and provide much better results than ChatGPT because of fine-tuning.
​
The PDF Chatbot takes in a single PDF and answers questions based on that document. You can choose between models from OpenAI, Anthropic, and Anyscale for the chatbot to use while answering questions.
The Feedback
“Myles used our documents and data to create a fantastic chatbot. The speed and accuracy of the chatbot was truly impressive."