Sancus: The AI-powered magnet to find the needle in a data haystack

Sancus empowers its users to significantly reduce the time they spend looking for the right answer. We do so by consolidating your firms’ scattered information into a vector knowledge database. Leveraging a combination of information retrieval techniques and LLMs, our modular platform enables the user to simply pose their questions via a chat interface and receive easily verifiable answers, streamlining the process of accessing critical information.

Our platform is already leveraged by the Dutch Ministry of Social Affairs and Employment to conduct research on hundreds of collective labor agreements in a matter of minutes.
KEy Metrics
relevance of generated answers, calculated in standard benchmark tests.
elapsed time per data retrieval, calculated for 100 files.
reduction of process integration time
The Challenge
Organisations today have vast amounts of unstructured textual and tabular data. Such heterogeneity poses multiple challenges in leveraging the firm’s knowledge efficiently and effectively.

Moreover, integrating and processing such data requires robust Natural Language Processing (NLP) techniques. Yet, these oftentimes carry their own potentially harmful biases, which limit their ability to leverage specialised knowledge. Therefore, custom approaches are needed for refining text extraction, cleaning, and normalisation, so that companies can ensure their data is well-utilised across various domains.
Our Solution
Sancus provides users with an intuitive interface for managing and interacting with their company’s knowledge base using natural language. The user-friendly interface is powered by:
  • Data ingestion process that prepares the documents to be used by the other modules. We are able to support a wide variety of data and documental formats (e.g., from emails to Notion pages) due to our use of the Langchain framework
  • Encoder which play a crucial role in transforming data into a meaningful representation that can be efficiently retrieved and utilised by downstream components. This enables effective aggregation and subsequent generation of responses, facilitating the system's ability to comprehend and respond to user queries comprehensively.
  • Vector database which saves the knowledge base in a database that keeps the semantic meaning of sections of the document. This allows the tool to find sections of the document that are semantically similar to the posed question.
  • Retriever which reranks the results from the vector database by using a pre-trained model like ADA to provide better answers to the questions.
  • LLM (e.g., GPT-3.5 turbo) that enables us to provide a clear and simple answer for the results found in the vector database.

Finally, Sancus is built in a modular way, which allows us to continuously monitor and upgrade both each tool’s performance, as well as its end-to-end flow. As such, we can also easily switch to new technologies as they appear on the market.
The Impact

Efficient Retrieval

Of tabular and text data that allows for quick and easy extraction of the answer to each user’s questions from the company’s knowledge base.

Centralised Knowledge Base

Ensuring the knowledge of an organisation is easily accessible and verifiable via the same platform.

Accurate Answers

Paired with overview on where the information was found to increase output confidence.
The Ministry of Social Affairs and Employment continuously monitors changes in collective labor agreements (CLAs) across different industries, through research. This meant that they had to manually comb through hundreds of CLAs to find answers to their research questions. This was a tedious and time-consuming process.
Sancus, with its natural language processing capabilities, enabled the Ministry to automate and augment the existing labour intensive process of searching and analysing CLAs. It also increased the overall accuracy, reduced the number of mistakes and improved the job satisfaction of the staff.
Sancus provides its users with an intuitive platform that simplifies managing and extracting insights from extensive knowledge bases using NLP techniques. Further, its modular design supports continuous improvement and easy integration of the latest developments. Already adopted by the Dutch Ministry of Social Affairs and Employment, our platform reduces research time, increases decision-making accuracy and streamlines the firm’s operations.

Curious to learn more?

Reach out to us!