Turning Unstructured Documents into Structured Information with Generative AI

Use case

A public institution in a Central European country faced a challenge in creating a specific report. The process involved manually analyzing thousands of contracts from the central register of contracts to extract data on IT service pricing, scope, and roles. This manual analysis was time-consuming and resource-intensive.

Solution

Cognexa proposed a Proof of Concept (PoC) project to demonstrate the potential of Large Language Models (LLMs) in automating data extraction from contracts. The project aimed to create a software solution that could extract relevant data points using LLMs and assess its accuracy by comparing results with existing manually obtained data.

Result

The PoC project successfully demonstrated the feasibility of using LLMs for data extraction from contracts. The implemented system could accurately identify contract types, relevant IT roles, rates per person-day, and map roles to generalized categories. This automation significantly reduced the manual effort required in report generation, enabling the institution to produce the specific report more efficiently and effectively.