AI & DATA ENGINEERING
Data Accuracy with LLM and Azure AI for Financial Company
The Client Project
They needed a solution that could handle the vast variability in their document formats. The company needed improvements in its existing setup to enhance data extraction accuracy and ensure scalability for future reports.
The Financial company supports numerous industries and leverages advanced data analytics to improve decision-making processes. In this case, their client was CTL Thompson. The client initially used Python, Azure AI Document Intelligence, SQLAlchemy, and OpenAI GPT-3.5 to extract technical data from these geological reports.
Their system hosted documents on Azure Doc Storage and aimed to automate extracting data, such as site identifiers, research outcomes, construction recommendations, and client details.
Requirements
Implement a better structured Langchain integration to optimize search and infer more relevant data points from the reports.
Automate the extraction process whenever new files are uploaded to Azure Doc Storage using Azure Functions, removing the need for manual interventions.
Replace the previous SQL-based search with Text Embeddings using Langchain for more accurate results.
Implement Asyncio and threading to reduce the time required for data extraction by allowing parallel processing, especially when dealing with large sets of documents.
Add better control over inference parameters, enabling easy optimization and fine-tuning for the model’s performance.
Our Implemented Service
Engineers added features to control LLM generation parameters, allowing easier optimization of the model and more accurate data extraction.
We Improved the structure of the instructions passed to the LLM to get more reliable and precise inference results and replaced the previous SQL-based search with Text Embeddings using Langchain to improve the relevance of search results and reduce irrelevant information.
Finally, we implemented Asyncio and threading to parallelize processes, reducing the time required for data extraction from the reports.
Transform your data extraction processes with our advanced AI-driven solutions. Get a Quote
Our Tech Stack
The Results
The newly implemented system provided the fintech company with highly configurable and efficient data extraction capabilities.
By optimizing the LLM process, reducing unnecessary data, and enhancing resource utilization, the client experienced much faster and more accurate results. This improvement is expected to lead to the successful completion of their project, allowing them to extract valuable insights from decades of geological data.
