1. About the Company
Mahle Metal Leve SA is a Brazilian auto parts company, headquartered in Mogi Guaçu (SP), specializing in the manufacture and sale of components for internal combustion engines and automotive filters. Founded in 1951, when Ernst Mahle established himself in Brazil as a founding partner of Metal Leve, the company has grown to become one of the leading references in the national and global automotive sector. It is a subsidiary of Mahle GmbH , a German multinational company headquartered in Stuttgart, considered one of the world's largest automotive suppliers. In Brazil, Mahle Metal Leve operates in both the original equipment manufacturer (OEM) and aftermarket sectors, with products sold in dozens of countries. The company is recognized for its focus on technological innovation and has had its own technology center in Brazil since 1978.
2. objective
Develop and deploy a Generative AI- based technical support chatbot , hosted on AWS, capable of accurately and scalably answering technical questions from mechanics and professionals in the automotive aftermarket with low operational costs — replacing the previous solution based on Pinecone and OpenAI, which had high latency, high costs, and performance limitations.
3. Problem to be Addressed
Mahle already had a technical chatbot solution for mechanics via WhatsApp, but faced critical challenges that compromised its effectiveness and viability:
• High latency and low performance: The previous model (Pinecone + Flowwise + OpenAI APIs) had difficulties with complex queries that required contextualization and information association.
• High operational cost: The solution did not generate direct revenue and needed to be optimized to reduce expenses.
• Unstructured data: The data sources (spreadsheets and technical documents) were not adequately structured to provide quality feeds to the AI.
• Limited scalability: The solution was not prepared for expansion to other Latin American countries, languages, or integration with systems such as Microsoft's CRM Dynamics.
• Dependence on third-party technologies: The previous architecture generated dependence on external suppliers with less control over costs and evolution.
4. AWS Solution and Resources Used
Flexa developed an intelligent chatbot based on Generative AI, with a modern and scalable architecture hosted on AWS, completely replacing the previous solution. The main components used were:
• Amazon Bedrock: Generative AI engine with models like Amazon Nova and Claude (Haiku/Sonnet), providing cost reductions of up to 80% compared to the previous solution.
• Amazon S3: Storage of technical knowledge bases (spreadsheets, PDFs, manuals, warranty documents), segmented by category (light, heavy, motorcycle, agricultural).
• AWS Glue + Amazon Athena: ETL pipeline for ingesting, processing, and incrementally updating technical databases in Parquet format.
• Amazon ECS Fargate: Application hosting with scalability, security, and high availability.
• AWS Lambda + API Gateway: Orchestration of relational and vector database queries, with serverless execution.
• Amazon RDS: Relational database for structured queries of parts and components.
• AWS Amplify: Publish the React front-end with automatic deployment via GitHub.
• Amazon CloudWatch + Zabbix: Infrastructure monitoring, logs, and solution observability.
• Amazon Bedrock Guardrails: Security controls for content filters, denied topics, and protection of sensitive data (PII).
• AgentFlow (Flowwise/Flowise): Low-code/no-code platform for orchestrating chatbot conversational flows.
5. Business Benefits Generated
• Reduced operational costs: Replacing OpenAI APIs with Amazon Nova resulted in an estimated reduction of up to 80% in inference costs.
• Scalable technical support: Mechanics and aftermarket professionals now have autonomous access to accurate technical information about parts, warranties, and applications.
• Response quality: After intensive refinement cycles, the chatbot achieved an accuracy rate of around 80% in automated tests.
• Global expansion: The architecture was designed for multilingual support and expansion into Europe, the US, and Latin America, with environments segregated by region.
• Autonomy of the Mahle technical team: Complete knowledge transfer, with delivery of the source code, AS-Built documentation, and onboarding of the internal team.
• Potential for new projects: The chatbot's success has opened doors for future initiatives, such as demand forecasting and expanding the solution to other areas of the company.
• Global recognition: The project was presented to Mahle's global board in Germany, where it was received positively and with interest in international replication.
• Satisfied customer: Project delivered in April 2026, with the client opting to continue with the operational phase with Flexa's support.
