A customer from the chemical industry had collected internal research over many decades, consisting of several hundred thousand research papers and articles. A system based on a traditional document search engine was already in use. However, the client's global research community was constrained by the limited capabilities of the traditional search engine. Therefore, the client wanted to achieve more accurate results and better usability by allowing researchers to formulate more complex search queries in natural language.
We have developed a chatbot-based RAG (Retrieval Augmented Generation) system that follows the standard RAG paradigm. The system uses the combination of text embeddings of research documents and a large language model to convert queries from researchers into an interactive document search. Source fidelity, which is essential for scientific work, is ensured by the system referring to the original section of the research article for each response. The chatbot processes text queries by using prompt flows, retrieval mechanisms and validation logic implemented by HMS to retrieve and present the most relevant document excerpts.
Depending on the researcher's search intent, the system selects an appropriate processing logic. To identify the best processing logics (prompt flows), we have established an automated scoring system.
Advantages of the tool:
The HMS-RAG system enables the client to access research data faster and more accurately, leading to improved decision-making processes and increased efficiency in research questions. HMS supported the development and operationalisation of this project, demonstrating the direct business value of LLMs in enterprise environments.
