Skip to content

The Modulai way of working

We believe that a well-reasoned and carefully prepared process is key to success

Contact us to hear more

Our way of working

Whether we work on our internal projects, together with a long-term partner or develop a pilot project with a startup, we believe that a well-reasoned and carefully prepared process is key to success. At Modulai, we’ve refined ours based on what we’ve learned from developing 100’s of models with 50 or so clients.

Identifying opportunities
Regardless of your organization’s level of AI maturity, we’ve something valuable to offer. For you who have just begun your journey, we research your organization’s needs carefully and work with various stakeholders to identify opportunities for increased efficiency, cost savings, and improved customer experience.  

Definition and pilot
Most of our clients have a clear idea of what they need from start. They’re entrepreneurs in need of a machine learning component for their product or a large multinational company that wants to elevate a specific business area with the help of AI. Here the challenge is to transform that business problem into a set of realistic operations – an approach – and move the concept to a pilot – a proof-of-value.

Development and integration
The final solution is developed and tuned when the final scope has been decided. We build with attention to business needs and value creation. We take great care in building scalable and easy-to-maintain software and, depending on the needs, work closely with the client’s tech team to build the surrounding software, and data pipelines and integrate them with their consuming systems.

Identify

Together with the client we ideate and identify business opportunities for AI impementation

Given that a client wants to assess their general maturity for ML implementation and need to identify opportunities for ML/AI in multiple areas, we set up interactive workshops with business people (to understand company needs and strategy) as well as with tech people (to understand technical maturity, data sources, and execution environment) to refine and assess ideas for further development. 

Define

When use case for ML exists, we turn it into a method and plan forward

This is the starting point for most projects. A prerequisite is that a distinct use case exists. We set the project goals in terms of model performance, execution time, and other constraints. User journey, intended output formats, and execution environment are taken into account.  We produce a detailed plan for the pilot stage as well as a high-level plan for the following development stages. Methodology, technical and scientific approaches, down to specific initial algorithms are decided. This step is executed in a combination of workshops with stakeholders and internal planning and scoping.

Pilot

We prove the value by analyzing data and developing models

In this stage, we aim at trying out the methodology, building an initial pipeline, training models, and assessing performance. We set up tools and services for project planning, code versioning, and project-related communication.
Results are delivered in the form of the code base, presentation of results, and recommendations for further development. We assess the feasibility of the project, both in terms of technical complexity and the value it would create. A detailed plan for the later stages of the project is decided upon and shared with the client.

Given that a client wants to assess their general maturity for ML implementation and need to identify opportunities for ML/AI in multiple areas, we set up interactive workshops with business people (to understand company needs and strategy) as well as with tech people (to understand technical maturity, data sources, and execution environment) to refine and assess ideas for further development. 

This is the starting point for most projects. A prerequisite is that a distinct use case exists. We set the project goals in terms of model performance, execution time, and other constraints. User journey, intended output formats, and execution environment are taken into account.  We produce a detailed plan for the pilot stage as well as a high-level plan for the following development stages. Methodology, technical and scientific approaches, down to specific initial algorithms are decided. This step is executed in a combination of workshops with stakeholders and internal planning and scoping.

In this stage, we aim at trying out the methodology, building an initial pipeline, training models, and assessing performance. We set up tools and services for project planning, code versioning, and project-related communication.
Results are delivered in the form of the code base, presentation of results, and recommendations for further development. We assess the feasibility of the project, both in terms of technical complexity and the value it would create. A detailed plan for the later stages of the project is decided upon and shared with the client.

Develop

Development of the final solution – value creation is key for success

Continuation of the pilot phase, but with data accessible in the production environment. Various trade-offs in terms of performance vs execution speed and complexity are taken into account – value creation is key. We build data pipelines, carefully train and validate models, and develop much of the inference code.

Integrate

Model hosting, inference service and integration with consuming system

In this stage, we put all the parts together and make sure that it runs smoothly. We develop the various micro-services needed for model hosting, execution of business logic, and integration with the client’s back-end, data sources, and front-end services. When this stage is finalized, the newly developed AI component becomes available for use by the client.

Maintain

We offer monitoring and maintainance of the final system solution

Models are monitored for drift and errors. Possible improvement opportunities are detected and implemented. When needed, models are retrained preprocessing and post-processing steps are adjusted. This step can be done by Modulai, either through a long-term maintenance agreement or by specific inquiries from the client. Often, the developed services are handed over to the client and they perform maintenance themselves with our occasional advice and supervision.

Continuation of the pilot phase, but with data accessible in the production environment. Various trade-offs in terms of performance vs execution speed and complexity are taken into account – value creation is key. We build data pipelines, carefully train and validate models, and develop much of the inference code.

In this stage, we put all the parts together and make sure that it runs smoothly. We develop the various micro-services needed for model hosting, execution of business logic, and integration with the client’s back-end, data sources, and front-end services. When this stage is finalized, the newly developed AI component becomes available for use by the client.

Models are monitored for drift and errors. Possible improvement opportunities are detected and implemented. When needed, models are retrained preprocessing and post-processing steps are adjusted. This step can be done by Modulai, either through a long-term maintenance agreement or by specific inquiries from the client. Often, the developed services are handed over to the client and they perform maintenance themselves with our occasional advice and supervision.

Do you wanna talk more about how we can help your company?

Contact us