General Information / format of event

The National Bank of Georgia continues to encourage data science popularization and invites people interested in modeling to Techsprint. The event will be held primarily virtually during two days. Participation in the Techsprint is expected in the form of teams. The Team can also consist of a single participant. We anticipate that teams will comprise up to five members performing different functions to develop the working solution. The teams will be formed on November 5 at the head office of the National Bank. Participants will be able to choose team members themselves. Participants must submit the application form at the following address.

Data scientists, machine learning engineers, or anyone interested in artificial intelligence who has modeling experience and are passionate about challenges are welcome to participate in the Techsprint.

Goal

The goal of the Techprint is to create an OCR solution that can read Georgian-language official documents in different formats (font, quality, mixed with other languages) with high accuracy and convert the given information into digital form. Participants will not be constrained in model or data selection or development process: they are allowed to train a new model, retrain existing models for a specific task, create data, use an existing one, and so on. Teams will need to demonstrate that they have a deployable and scalable solution to the problem statement. The team should provide a brief documentation of model, where solution will be explained.

Evaluation

Participants will be evaluated on two aspects:

• Adequacy and soundness of the solution’s methodology
• Accuracy of forecast

The accuracy of the forecast will be assessed on the synthetic test set. Test data will be distributed to participants after completion of the model development phase. Teams will have some time to evaluate the model results in the test set and to estimate the error. The teams would subsequently submit their developed solution methodology and test result in the form of presentation to be assessed by a panel of experts, after which the winning team will be identified.