Machine Learning Solutions

Building customized and robust machine learning software: from development and modeling to production.

At dida we develop versatile machine learning software that easily integrates into existing IT infrastructure. Our tailor-made solutions aim to address a wide range of applications from numerous and diverse industries, covering all fields of machine learning.

Customized AI software

Often blackbox solutions do not achieve the required accuracy - due to specific data scenarios or a combination of very challenging problems. With our problem-specific approach, we design AI software that is tailored to the respective use case.



It’s all about the details

Developing complex AI solutions requires going deep into technical and mathematical details. In practice, it often turns out that only a combination of novel machine learning techniques and creative approaches from more classical branches of mathematics and engineering leads to outstanding results.

We are a team of experts

Our interdisciplinary team has a strong background in both theory and practice - many members hold a PhD in mathematics or physics and all of them are well experienced with state-of-the-art machine learning techniques.

Focus on production software

We have provenly escaped the PoC-to-application gap and have deployed most of our software in production systems, where we care about scalability, maintainability and MLOps.

Development Process

The process of developing a successful machine learning software is complex and requires multiple steps, which we shall briefly summarize in the sequel.

Setting main objectives

In first informal exchanges, we aim to fully understand the problem at hand. This usually works best when respective data is available. We can then define clear objectives and provide a detailed project proposal.

Feasibility study

In a feasibility study, usually lasting 2-3 months, we aim to provide a proof of concept of the desired goal. It is usually constructive to define a clear evaluation metric that measures the success of this phase.

Development of production software

Given that the feasibility study was positive, we then aim to advance the software to a production level, usually lasting 3-9 months. This includes improving model performance and generalization to all relevant production scenarios as well as robust deployment pipelines and MLOps.

Maintenance and additional features

We offer further maintenance once the software is running in production, in particular monitoring of model performance or retraining in shifted data settings. Further feature requests can also be incorporated.