Training and software development to support vineyard management

We helped grow an SME's machine learning expertise through the exploration of AI-based solutions for its app.

"We had a very positive experience during the project with EPCC, and the service very well complemented our skillset and knowledge with the expertise we were lacking, helping us not only to build new features with technologies on a much more advanced level than we could do alone but also to obtain relevant skills that we will be able to use in the future,"

Matic Ĺ erc CEO, ELMIBIT

ELMIBIT is a Slovenian-based SME developing and distributing digital agriculture software solutions, including the eVineyard vineyard management software. Winemakers worldwide use the eVineyard application to make more informed decisions on the next vineyard activities, plan work, manage vineyard teams, and lower production costs. 

The challenge

ELMIBIT came to EPCC with the primary aim of developing and improving their in-house skills in modern AI programming techniques. To ensure good value, EPCC and ELMIBIT identified relevant R&D software development projects that could be solved with simple AI methods, and demonstrated the implementation of these methods to train eVineyard developers in the use of AI techniques. This approach ensured that the training was relevant to the company's day-to-day operations, and that ELMIBIT would gain valuable R&D insights as well as the required training. 

ELMIBIT and EPCC identified two problems ideally suited for this approach:

  • ELMIBIT wanted to add a functionality to its app that could predict the expected price of a grape harvest based on the weather and harvest yield in a given European region. ELMIBIT has access to public datasets of historic information on grape harvest yields, prices, and historic weather data for these regions.
  • ELMIBIT has developed a highly-regarded model for estimating the ideal time for picking grapes on a vineyard-by-vineyard basis. This method works very well for short-term predictions close to the picking date, but is less reliable earlier on in the growing process. ELMIBIT asked EPCC to explore the potential of AI methods for making longer-term predictions.

Our solution

EPCC worked closely with the eVineyard software team to develop an approach that would maximise training and learning opportunities while also solving the two problems defined above.

We designed well-documented and heavily-commented Jupyter Notebooks that would enable the eVineyard software developers to go through, understand, and recreate the work that EPCC had done. This understanding was reinforced through a series of online workshops where EPCC explained in-depth the reasoning behind development choices and answered any questions that the eVineyard developers had.

While doing this, EPCC was also able to deliver on both problems. We developed a simple multilayer perceptron model to predict the harvest price based on the weather and yield of the grape harvest in a specific region. Additionally, we developed a statistical autoregression that can predict almost exactly the ideal time to pick the grapes up to three weeks ahead of time, and can predict the ideal picking time up to within a week up to six weeks ahead of time based on a region's historic weather patterns. This information could prove of great value to users of eVineyard, giving them an edge when hiring grape pickers for harvests before other winemakers and enable them to better prepare for the harvest.

Contact us

To explore how EPCC's expertise can support your business goals, please contact our Commercial Manager, Julien Sindt.