Urgent project rescue.

Case study

Urgent project rescue.

Re-inventing the faulty solution for managing machine learning models.

Challenge

Salvaging the solution for managing machine learning models partially-delivered by another provider and extending its scope to ensure friction-free experience for a Telecom company.

Solutions

Mid-project intervention to reclaim the potential value of a set of applications being provided by a top consulting company, whilst minimising budget increase and delays. We alleviated the mismatch between the solution and company’s infrastructure and introduced necessary adjustments to ensure compatibility. Once the potential was salvaged, we enriched the solution by building add-ons and adding features to provide the missing functionalities, which were initially expected.

Technology & Tools

Python

scikitlearn

pandas

Spark

Airflow

MLFlow

Client

A large enterprise leader in the telecommunications industry across the Nordic-Baltic region. Aiming at improving their ML to seamlessly enable the digital societies via providing essential digital infrastructure and services.

Opportunity

Increasing functionality afforded by the ML models in the digital-powered era leaves companies with complex features that provide vast potential for capitalising on and increasing revenue streams.

To tap into and amplify the effectiveness of such solutions, a management device that merits friction free usage for the data scientists and provides reliable results is required. Leaning into the opportunity this provides, the telecom company invested in the infrastructure of their MLOps, which were fragmented and required tuning to enable automatisation, joint management and synergies.

However the consultancy they have first hired to realise the value of their ML management, encountered issues with the mismatch between the company and their proposed solution, as well as, did not manage to ensure the assumed functionalities within the code. The predicament called for quick support from specialists, with expertise to re-architect and extend the solution for the perfect fit and delivery.

Delivery

The intervention started with a comprehensive audit of the bottlenecks hindering the project’s critical line. Working together with the consultancy team to truly understand the framework and its intended potential, our experts mapped the gaps, moved on to verifying the relevancy of functionalities with employees’ to maximise the outcome for current and potential needs. Next they acclimated the management system, by successfully obviating the mismatch causes and re-architecturing the solution for the real daily needs. In the next stage, because only the core elements of the initial device were relevant, the missing features such as: the lack of resource management system or nonexistent ability to deploy the models to an existing environment; were implemented from zero. Moving onto the clean up, the refactoring and fixing of pain and problem points helped remove the barrier between the client and their effectiveness. Lastly the solution was optimised to encompass all models and enable extending their scopes.

Effect

The solution was perfectly recrafted towards the needs and environment of the company and the data scientists were trained for ease of navigation, to truly unlock the potential of ML. It allowed Data Scientists to focus on building the model while the model maintenance was automated to a great extent and delegated to Engineers dedicated to MLOps activities. The initial solution contained 12 machine learning models and the simplicity of implementation allowed the client to plan for migration of another 40 models to the new solution.

Testimonial

Head of Data Services, Telecommunication Company