Answering Technical Queries in Less than a Minute with Custom Chatbot

German company Everllence, part of Volkswagen Group, is a leading solutions provider for the maritime, power, and industrial sectors. Maritime engineers travel the globe, working with various engine types. They’re often in the middle of the ocean when they must troubleshoot problems, a process that can take hours. “The idea of a chatbot came about to give instantaneous answers so customers can solve their problems quickly,” says Gregory Puckett, chief digital officer at Everllence.

So, Everllence developed a chatbot that integrates with Everllence CEON, its cloud-based system hosted on Amazon Web Services (AWS). To tap into a broad availability of large language models, Everllence used Amazon Bedrock, which provides developers with a simple way to build and scale generative AI applications with foundation models. “Amazon Bedrock makes it easy to switch and test different models,” says Christian Hendricks, principal data scientist at Everllence. “We really love it.”

When maritime engineers encounter challenges, they can ask chatbot questions about engine performance, maintenance procedures, and troubleshooting steps using natural language. Everllence provides engineers with relevant, accurate responses by using Anthropic’s Claude models in Amazon Bedrock.

Anthropic’s Claude models are designed with safety guardrails that can detect abusive language and mitigate hallucinations. “We are focused on providing as factual information as possible with a high level of security,” says Carsten Hounsgaard, product owner of the AI Lab at Everllence.

Everllence powers its chatbot on 5 million technical documents and real-time equipment telemetry data from 1,778 connected sites, and it can respond to complex queries within 1 minute or less.

If the chatbot cannot find reliable information to answer a query, it explicitly states this rather than providing potentially incorrect information. This approach has built trust with users. "That is the value of the system — we are sure that the answers come from validated sources," says Hendricks.

The chatbot has sparked an AI transformation at Everllence. The company is enhancing the solution to answer more complex questions that require multiple lookups across documentation sources. Additionally, Everllence plans to expand the chatbot to new use cases, such as automating daily equipment reports and integrating with spare parts ordering systems.

Everllence is also developing features that will help internal teams perform their own data mining without relying on data scientists. After completing an internal release and gathering feedback, the company plans to roll out these advanced capabilities to external users. "We see this going beyond simple answers to a much more enhanced self-service portal for our customers," says Puckett.