Traffic_Regulations_Chatbot

Project overview

• Developed a chatbot application to help users to clarify questions about Mexico City’s traffic regulations
• Leveraged LangChain, AWS Bedrock, and OpenSearch for the backend infrastructure.

• Implemented the Retrieval-Augmented Generation methodology, using Titan for embeddings and Claude Haiku as the Language Model.

  

</p>

Part 1: Creating a Vector Database

• Implemented AWS OpenSearch: Utilized the boto3 library to interact with the AWS OpenSearch service, establishing a secure and scalable vector database.
• Why AWS OpenSearch? Selected for its efficient vector storage and retrieval capabilities, seamless integration with AWS services, and advanced security features—making it an optimal choice for this chatbot application.
• You can see the full and detailed script used to create this DB HERE

Part 2: Backend

• PDF Processing: Developed a function to load PDF documents and split them into smaller, manageable chunks.
• Embedding Generation: Configured AWS Bedrock with Titan to generate embeddings from the processed PDFs.
• Vector Storage: Stored the generated embeddings in an OpenSearch index, functioning as the vector database.
• QA Chain Setup: Combined the vector database as retrival with the LLM, creating a pipeline that takes a user query, retrieves relevant document sections, and generates a concise answer.
• You can see the full and detailed backend script HERE

Part 3: Frontend

• User Interface: Developed a user-friendly interface using Streamlit, allowing users to easily interact with the chatbot.
• Backend Integration: Seamlessly integrated backend functionalities into the frontend, enabling real-time user query submission and response generation.
• Performance Optimization: Implemented session management to ensure the vector index is only initialized once, optimizing performance and user experience.
• You can see the full and detailed frontend script HERE

Demo

• First we initialize the app

• We ask the question

• We get the answer

Conclusions

• Successful Implementation: Developed an effective chatbot application for clarifying Mexico City’s traffic regulations, demonstrating a solid application of LangChain, AWS Bedrock, and OpenSearch.
• Methodology Efficiency: The use of RAG with Titan embeddings and Claude Haiku as the LLM proved to be an efficient approach, providing users with precise and contextual answers.