CSAS: Chat with your knowledge base​

Challenge

Motivation to change

Ceska Sporitelna possesses large amounts of documents like guidelines, product documents and similar.​

Ceska Sporitelna hired ableneo to develop a knowledge base search application on top of these documents in form of a chatbot.​

The application is split to two separate workflows.​

First is the ingestion, responsible for inserting new documents, unlearning outdated ones and keeping metadata related to the knowledge (owners, groups allowed to see the data, etc.).​

Second is the chatbot application, where the input is the user’s question and the output is the response combining all the sources in a single well-structured answer, citing all the sources.​

Solution

Change delivery

CS’s initial vision of the application was a native Microsoft Azure based solution, which scored around 64%. Since this was not fulfilling the KPIs, ableneo utilized its expertise in the field of Natural Language Processing (NLP) and Language Models (LM) to increase the success rate to 86.4%.​

Tools & means

• Microsoft Azure​
• Azure OpenAI GPT4​
• Azure AI Search​
• LangChain​
• Natural Language Processing​
• Prompt Engineering

Outcomes

Change outcomes

We have successfully delivered a chatbot application correctly responding to 86.4% of the questions.

Used services