The AIKA stories

The AIKA ecosystem reference stories are PoC-type mini-projects, where our team has generally used pre-trained AI models, analytics or standard business intelligence visualization to solve a problem statement given by the pilot company. The company provides the data and a hypothesis on how the data should predict or correlate with the expected results, which should contribute to the business goals discussed between the company and AIKA team. AIKA technical specialists have then attempted to reach a reasonable accuracy and validated the hypothesis. The results have varied between a report presentation, a lightweight demo-tool or an infographic for the company, summarizing whether they could use AI or related technologies in their everyday operations in the context of the previously stated business problem.

 

Esko Systems Oy

Esko Systems Oy is a Finnish non-profit in-house company specializing in the development of client and patient information systems for the social and healthcare sectors. Founded in 2019, Esko Systems operates from offices in Oulu, Rovaniemi, and Kajaani, and is owned by its public sector customers, including several well-being services counties and IT service providers. The company`s flagship solution, Esko EHR, is a comprehensive patient information system designed to meet the evolving needs of Finnish healthcare, integrating functionalities across specialized care, primary care, and patient management.

Esko Systems is known for its agile development model, working closely with end-users to ensure that its systems are reliable, cost-effective, and tailored to real-world healthcare workflows. The company’s mission is to enhance patient safety and continuity of care by enabling seamless access to information across healthcare domains.

In collaboration with the AIKA Ecosystem and Health Hub Finland, Esko Systems participated in a pilot project focused on benchmarking large language models (LLMs) for medical dialogue support. The goal was to evaluate how different LLMs trained on various datasets could interpret and respond to medical terminology and conversational nuances in clinical settings.

Due to privacy constraints, the benchmarking was conducted using semi-synthetic data representative of actual medical interactions. The AIKA team designed a controlled testing environment to compare model performance across several dimensions, including terminology recognition, contextual understanding, and conversational clarity.

The pilot provided Esko Systems with actionable insights into the strengths and limitations of different LLMs, as well as guidance on prompt engineering strategies that could improve AI responsiveness in healthcare contexts. This proof-of-concept demonstrated the potential of AI to support clinical decision-making and patient communication, while respecting the stringent data protection requirements of the healthcare sector.

 

Critical Force Oy

 

Critical Force Oy is a pioneering Finnish game development company specializing in mobile multiplayer shooter games, with headquarters in Kajaani and additional operations in Helsinki. Founded in 2012, Critical Force is best known for its flagship title, Critical Ops, which has achieved over 145 million downloads worldwide. Critical Force employs around 30 professionals and is recognized for its vibrant, international team and commitment to innovation in the gaming sector.

In collaboration with the AIKA Ecosystem, the Critical Force team explored the use of machine learning -based classification algorithms to enhance player engagement and retention. The project focused on two key challenges: classifying gamer profiles and predicting player churn.

The AIKA team developed a predictive model, a proof-of-concept, that analyzed player data to identify both users at risk of churning and those who were highly engaged. This proof-of-concept provided Critical Force with valuable insights on how to further develop their customer engagement strategies.

 

 

Granlund Oy

Granlund Oy is a leading Finnish expert group in the real estate and construction sectors, with over 1,500 professionals across 30 locations in Finland. The company focuses on improving the functionality, intelligence, and sustainability of buildings through services in technical design, energy and environmental consulting, software development, and lifecycle management. Granlund’s mission is to build a smarter and more sustainable future by leveraging digital solutions and close collaboration with stakeholders.

In collaboration with Granlund’s Kainuu regional branch, the AIKA Ecosystem team explored opportunities to streamline a repetitive and time-consuming task: classifying building ventilation equipment in Excel sheets. This manual process involved assigning specific codes based on a set of technical and contextual rules, which was necessary for integrating the data into downstream systems used in property maintenance and management.

To demonstrate the potential of AI in solving such operational challenges, the team developed a proof-of-concept combining rule-based logic with AI-powered text interpretation. The component performed as expected and appeared to offer notable time savings. Based on these promising results, Granlund proceeded with implementing the solution in their workflow, where it has shown potential to save several working hours per week.

 

 

Iiwari Tracking Solutions Oy

Iiwari Tracking Solutions Oy is focused on providing indoor positioning systems, which require context-aware planning. This work is manual and time-consuming, especially in multi-room customer environments, such as hospitals, office spaces or daycare centers. As a technology company, Iiwari has the best know-how in improving their solution and service platform, so an AI-assisted sales and planning tool seemed to provide them with the best business value if proven accurate and flexible enough.  

Solution 

With the goal of providing their resellers and customers with a suitable cost-estimation and planning tool, the pilot was divided into two phases:  

1. Read in, process and recognize objects from customer floorplan documents. 
2.  Automatically place the main and child signal anchors to the floorplan.  

AIKA technical specialists explored pre-trained neural network models available for this purpose. The team decided together with Iiwari representatives to leave special cases, such as sports venues, out of the scope of the pilot due to the workload being much higher in cases of hospitals and offices. An approach relying on penalty score methodology was used in the second phase, but choosing the best way forward in it requires further studies and development. 

Results 

The pilot has been successful in proving the value of AI-assisted sales and planning for Iiwari who are therefore keen to explore production implementation, as well. 

 

“The AIKA researchers helped us in researching, innovating, evaluating and experimenting novel methods for solving real business scaling problem. Next step will be brining the research results to the production. We are happy for the co-operation and support received.” Esa Viljamaa, COO Iiwari

 

 

Kemet Electronics Oy (Yageo Corporation Inc) 

Kemet Electronics Oy is part of a global capacitor manufacturer, Yageo, but have rather specific specialization niche in Finland and are one of the biggest employers in the Kainuu region. The AIKA team ended up in discussions with Kemet late 2023 at our local AI workshop in Suomussalmi where several R&D and production-related possibilities with AI were discussed. The Kemet Finland factory is rather capable in terms of their manufacturing quality, but as the industry is increasingly competitive and the feasibility of available data seemed most promising, the teams decided to go ahead with a predictive maintenance business case. The goal for the case was simply being able to plan ahead for maintenance breaks and avoid downtime.  

Solution 
At first, AIKA team explored only maintenance data but after some visualization exercises, it was clear that more production volume -related datasets were needed to be able to anticipate component wear in capacitor injection machinery. There seemed to be some variation in the data between high volume producing machines and low volume ones and not always clear correlations in scrap production rate and the maintenance event. Getting further insight required some feature engineering for remaining production time and remaining products based on the historical values for the nozzles and their performance. The team changed the approaches between rough categorization of the machines and individual predictions but ended up with a normalized model that would perform well with typical production sets. Outliers and very low volume production cycles proved therefore problematic to predict. 

Results 

AIKA technical team prepared a model and went through several validation meetings for prediction accuracy. The model was implemented in a tool that could be used through Excel and was tested during Q12025. The exercise as a whole gave Kemet and AIKA experts valuable understanding of external factors influencing production equipment, as well as the structure and cohesion of data required for successful analysis. Therefore, it turned out that production capacity and maintenance events alone were insufficient to predict equipment failure. Kemet will focus on additional data points in their production and the teams are looking forward to further collaboration in the future. 

 

“The pilot study with AIKA helped us understand the foundations required for a successful predictive analytics or artificial intelligence utilization in our operations. Lessons learned from this concept will be used in further planning of data collection and data source unification efforts in the future.”   Janne Kinnunen, Managing Director, KEMET Electronics Oy

 

 

 

 

 

Kuhmo Oy

Kuhmo Oy is a sawmill company based in Kainuu and operating internationally. As for any company in the industry, classification of the incoming material is important for them for downstream quality. The quality of the logs is crucial, since a wrong decision early on may mean expensive concessions via customer refund claims. Image classification was selected as suitable data for an AI pilot if AIKA team could recognize certain defects and features from the logs to provide the company with an additional quality dimension to support downstream planning actions. 

Solution 

The first phase of the pilot was to pre-process the images to a suitable format for the AI algorithm. Various pre-processing methods were explored also based on the feedback from the Kuhmo Oy experts. The AIKA team did extensive research in the field of quality categorization in wood processing and employed a pre-trained algorithm to be further finetuned for their purpose.  

During multiple labeling and training iterations, it became clear that in order to successfully recognize a certain specific defect, large amounts of expert-labeled data were needed, as well as many iterations to assess the improvements. The accuracy of the model varied based on the representation of defects in the training datasets, but an improvement could be observed after each iteration. 

 

Results 

The PoC is being concluded in feasibility validation via a lightweight AI-quality sorting application developed by the AIKA team. The upward trend in accuracy will either prompt additional labeling iterations or “opt for defect” in uncertain scenarios for production use.  

 

Planray Oy

Planray manufactures and sells industrial heat tracing equipment and related monitoring software. They have also been a speaker at KAMK’s and other regional events regarding the importance of analytics in measurement technology. As a high technology SME and with their forward-thinking mindset, it became clear in early discussions that Planray had several suitable use cases for pilots and also good topics for academic research. At the moment, they analyze customers’ data manually on request to gain and share insights and it became a task for the AIKA team to try to formalize some of these insights into automated functions.  

Solution 

Although the pilot is still ongoing, early collaboration has already provided valuable insights. After the initial meetings about priorities and assumptions, the teams took a test customer database for inspection. At first, the assignment for AIKA was broadly to examine the data and look for correlations and outliers, as well as notable events before alarm triggers. The technical specialists of AIKA created visualizations from the dataset and additional meta-data was provided. The teams then organized a workshop to prioritize a few use cases for universal predictive maintenance purposes.  

Results 

After the initial data visualization, a workshop was organized between the AIKA team and Planray experts. From this workshop, earlier use cases were narrowed down to three, from which two most straightforward predictions were selected. As Planray offers analytics dashboards as part of their services, it is likely that the pilot will end up in implementation consideration when successful. 

 

Fincet Oy / Tool4Pro 

Fincet Oy is a Kainuu-based company providing maintenance shutdown management software, Tool4Pro, and related consultation. The entrepreneur behind the company has long experience from both mining and process plant maintenance operations, which fueled the source of potential insights in maintenance shutdown management. The problems in the area of managing successful maintenance shutdowns seemed to be numerous ranging from disconnected pen and paper & worksheet planning to lack of real-time maintenance systems, but despite being able to recognize potential issues, it was challenging to formalize the process in a programmatic manner.  

Solution 

The purpose of the pilot was to examine the data viability for predictive models in maintenance workloads and support planning efforts, as well as to highlight any conflicts of resources being allocated. It quickly became clear that solely from the maintenance event planning data, it was not possible to predict issues that might be related to resource allocation in different operations or some machine-specific, or logistical issue that could cause additional delays in the ramp-up. Although there may be various benefits in generative AI for easier maintenance operations, operational prediction looks to be more problematic due to scattered data sources.  

Results 

The team prepared rudimentary correlations between planned and real maintenance time, as well as repeated maintenance events but the dataset was too limited to provide reliable predictive models. For further evaluation, use of maintenance management system data was suggested to complement event occurrence which could then be used to better evaluate dependencies between the machinery, resources and maintenance shutdown planning.  

 

Would you like to add your story? Contact us now!

 

Head of AIKA ecosystem DIH & Data Nexus Solutions Lab

Anas Al Natsheh

+358 44 7101228 /+358 40 5849598

anas.alnatsheh@kamk.fi

www.aikaecosystem.com

www.kamk.fi

 

 

Customer Manager, CSC

Mikko Kerttula