End of Studies Internship - Optimizing Small MLLMs for Invoice Data Extraction - R&D

About Medius
Medius is a leading global provider of cloud-based spend management solutions, helping organizations drive their business forward by enabling best-in-class process efficiency, cost-savings, and greater financial control. 

Our modular spend management suite includes market leading solutions for payments, strategic sourcing, contract management, procurement, accounts payable automation, and supplier management as well as data insight tools bringing control, compliance and cost savings throughout the entire source-to-pay process.

Thanks to an easy cloud deployment, intuitive user interface, and built-in best practices, our customers experience tangible savings right from the start in addition to long-term ROI success. With Medius, spend is simply managed and savings are quickly realized by our customers.

Founded in 2001 and celebrating 20 years in business, Medius is owned by the global investment firm Marlin Equity Partners. In 2019 Medius acquired Wax Digital in The UK, followed by further acquisition of OnPay Solutions in USA to become a powerhouse in spend management solutions. The group has over 550 employees and offices in Sweden (HQ), the United States, the United Kingdom, Australia, Denmark, Norway, the Netherlands, Canada and Poland.

As a Gartner Magic Quadrant Visionary company, Medius continues investment in and execution of a substantial growth plan to reinforce Medius’ position as a global leader in the market. With a dedicated Medius Pay sales team, we extend our reach even further.

We have a culture of constantly challenging ourselves and our colleagues and appreciating an ethos of open feedback to support alignment and collaboration towards our common goals.  Add to this strong leadership from an experienced global executive management team and you'll find a terrific opportunity to join Medius and truly make a difference.  

If you are ready to challenge yourself in a business with incredible momentum and rapid growth; in a company leading the change – not following it – then come and join us here at Medius.

For more info: www.medius.com

Description: 

The goal of this internship is to evaluate and optimize a small Multi-Modal Large Language Model (MLLM), such as OpenBMB/MiniCPM-V 2.6, for data extraction from invoices, and compare its performance—accuracy, speed, and cost—against larger state-of-the-art models like GPT/OpenAI. The project will focus on understanding how smaller, mobile-optimized models can outperform larger models on tasks like OCR processing while being more efficient.

The selected candidate will work within the Medius Capture squad on:

• Evaluate the performance of MiniCPM-V 2.6 on data extraction from invoices, comparing it to models like GPT.
• Benchmark MiniCPM-V 2.6 using proprietary invoice datasets to assess its accuracy, speed, and efficiency.
• Conduct a comparative analysis of the results against other leading models, such as GPT, to identify strengths and weaknesses.
• Optimize the MiniCPM-V 2.6 model by tweaking its parameters and architecture to improve performance.
• Implement the optimized model as a data capture service for real-world invoice processing scenarios.
• Develop a case study to showcase the practical applications and benefits of MiniCPM-V 2.6 in real-world environments.
• Document the findings and provide a detailed performance report, including insights on deploying advanced MLLMs for cloud-based applications.
• Present the results and insights to contribute to the development of high-performance, accessible open-source MLLMs.

Technologies & environment: .Net, Payton, Azure Open AI, Large Language Models, prompt engineering, git.

Please send your resume in english (resumes in french will be automatically rejected)