Ladamerica

Development of an intelligent chatbot for consulting vehicle regulations

Retrieval-Augmented Generation (RAG)

Conversational AI

Public Sector Innovation

The Challenge

The system centralizes vehicle management regulations and procedures in Uruguay.

However, the information was scattered across multiple PDF documents (legal regulations, instructions, and circulars).

This made it difficult to quickly access clear answers to frequent questions from both users and officials.

Manual searches were inefficient and time-consuming.

As a result, the system posed a high operational burden and increased the risk of errors.

How did we do it?

We designed and implemented an AI-powered chatbot using Retrieval-Augmented Generation (RAG) architecture.

Semantic indexing with FAISS for three types of documents, using chunking tailored to each type.

Optimized vector embeddings (OpenAI) for accurate retrieval.

Answer generation with GPT-4 Turbo, including token control, conversational memory, and safe fallback handling.

A REST API built with FastAPI and a web interface integrated via Streamlit.

Containerization with Docker to simplify deployment.

Natural language access to regulations was successfully automated, improving response efficiency and reducing the need for human intervention.

The system can be easily integrated into web portals or internal applications.

It supports scalability by document type or region.

Response time for frequent queries was significantly reduced.

Accuracy and consistency of answers increased.

The system is easy to maintain, with a clear separation of logic, model, and embedded documents.

Internal resource usage was optimized by reducing time spent on maintenance and deployment.

The system is largely self-configurable, allowing users to adjust the service to their needs.