How Gen AI scaled knowledge access for 400+ automotive dealerships

The client wanted to revamp their existing platform to enable quick and seamless extraction of insights from diverse documents like PDFs and newsletters. They also wanted to build a mechanism for query classification and sentiment analysis which could enable continuous improvement. 
They sought to build a platform that offered 24*7 information access for dealers without long waiting times and high operational expenses. And multi-role support for both dealers and home office users, ensuring fast, accurate responses through an AI-powered chatbot. 

Nagarro has built a cost-efficient, serverless platform on AWS to deliver intelligent document processing and a global chatbot. We use Amazon Textract to extract data from PDFs and Amazon Bedrock to standardize tables and layouts.

The platform stores processed content in Amazon OpenSearch, using hybrid search with semantic and keyword matching, metadata filters, and precomputed embeddings for faster results. It saves queries and feedback in Amazon DynamoDB and uses Amazon Comprehend or Bedrock LLM to analyze sentiment and group similar queries for continuous improvement.

The chatbot runs on AWS CloudFront and S3 for global reach, secured by a Global Authentication System with role-based filtering for personalized responses. It uses short, context-rich prompts, time and document filters, and fuzzy matching to reduce errors and deliver accurate answers.

Caching common answers, batching document uploads, and fine-tuning search settings speed up performance and keep response times low.

The solution delivered better customer experience by enabling natural language Q&A with transparent, citation-based answers. It captures queries and feedback to uncover trends and sentiments, helping teams continuously improve responses.

Role-based filtering ensures relevance for different user groups and asynchronous design eliminates timeouts and reduces operational costs, making the system highly scalable.

Meanwhile, smarter search with metadata filters and hybrid techniques improve accuracy, reducing irrelevant results, and boosting confidence in answers. The reprocessing pipeline further enhances data quality by removing unnecessary content and preserving critical information, ensuring reliable document processing.