Data Science capstone projects batch #27

We want to take a moment to celebrate all the incredible students who joined us in September and gave their absolute best to complete the course and capstone projects.

In just three short months, the talented Data Science enthusiasts from Batch #27 in Zurich, alongside the accomplished 6th cohort from Munich, took on a diverse range of ambitious projects. Their exceptional skills, passion, and dedication were on full display every step of the way.

We are thankful to HP for their invaluable support, providing us with cutting-edge Z by HP workstations that empowered our students to achieve even more and contribute to their success.

This exciting journey is one we're thrilled to share with you — witness the power of data science in action as our students push boundaries, uncover powerful insights, and create meaningful impact.

Optimizing Packaging with Machine Learning

Students: Simon Püschel, Anahy Santiago

Best Secret, a prominent name in luxury fashion retail, has undertaken a project to enhance its package selection process through predictive modeling. With this initiative, the goal is to support packers in choosing the optimal box size swiftly and accurately, boosting both the speed and consistency of packing operations. 

Currently, Best Secret’s package selection process depends on experienced packers who determine box sizes based on the types and quantities of items in each order. However, this approach can be time-consuming, especially for seasonal workers who lack intuitive experience. 

Simon Püschel and Anahy Santiago aimed to develop a machine learning model to quickly and reliably predict the optimal package size for each order, streamlining the process. The task involved 11 highly imbalanced classes, sometimes of very similar size, adding an extra layer of complexity.

Without exact volume and weight data for individual items, the team developed methods to approximate these values. By analyzing single-item orders, they calculated average volumes for different product categories, providing the model with crucial volume estimates. For weight approximation, they used average weights across categories, which served as useful reference points to support accurate package size predictions. 

To account for item flexibility, which could influence packing decisions, the team introduced a basic classification system. Items with clear identifiers were labeled as “soft” or “hard.” For less obvious cases, a model analyzed product names to classify items based on general patterns. The model ultimately used core product data, such as volume, weight, and quantity, to improve the accuracy of package size predictions.

Model Selection and Performance: The team evaluated multiple machine learning models to identify the best approach for this task. Gradient boosting algorithms demonstrated the highest accuracy at 64%, outperforming models like Random Forest and Neural Networks. To further improve predictions, the team reframed the problem as a regression task, incorporating ensemble learning techniques and testing a range of regression models.

While initial results are promising, the team has identified several areas for further refinement. Developing a more detailed product categorization could help capture subtle differences in item characteristics, enhancing classification accuracy. Integrating actual weights for each product would improve model precision, and real-world testing in a live packing environment would provide valuable insights for adjustments. Thanks to the support and resources provided by Best Secret, the team has laid a strong foundation for a more efficient packaging process. As they continue refining the model, predictive modeling is set to play an increasingly essential role in optimizing packing operations. 

 

AI Coach – Your Virtual Assistant for Back Pain Self-Management

Students: Bing Huang, Martin Itten, Srinivas Reddy Burigari

Do you struggle with back pain? If so, you are not alone! A staggering 46% of employees report experiencing back pain, resulting in 50% more sick days compared to their peers who are pain-free. Back pain isn’t just a personal challenge - it’s a widespread issue that affects productivity and well-being. Healactively is on a mission to tackle the back pain epidemic with a science-based, personalized program of back-strengthening exercises and physiotherapy insights - all delivered through a user-friendly mobile app.

Constructor Academy data science graduates Srinivas Reddy Burigari, Martin Itten and Bing Huang introduce a Healactively AI Coach to guide users on their journey, answering questions, providing emotional support, and building healthy habits for lasting back health.

They leveraged cutting-edge techniques like Retrieval-Augmented Generation (RAG) and the latest OpenAI models to tackle this challenge. As large language models (LLM), such as ChatGPT, do not have prior knowledge about physiotherapy or the exercise programs in the user's training program, the team uses RAG to bridge the gap. This approach retrieves relevant information from the company’s knowledge base and integrates it with the LLM, enabling it to deliver accurate, personalized responses. Additionally, the team manages chat history effectively, ensuring that user interactions are context-aware, seamless, and consistently helpful throughout their back health journey.

 

The AI Coach is powered by four specialized agents, each designed to enhance the user experience and provide comprehensive support:

Agent 1: Customer Support
This agent handles general queries from users, covering topics like subscriptions, memberships, and exercise programs, etc. Whether a B2B or B2C user, it tailors responses to meet the user's unique needs. Plus, it adjusts its tone based on the user’s persona, ensuring a personalized and engaging interaction.

Agent 2: Virtual Physio Expert
Need physiotherapy advice? The AI coach is here to help! Using user health data and pain levels, it provides customized recommendations to support the recovery and help users build a stronger, healthier back.

Agent 3: Habit Planner
The Habit Planner helps users build sustainable habits around their exercise routines. By analyzing personalized exercise plans, it identifies convenient times or triggers to seamlessly integrate back health exercises into daily schedules.

Agent 4: Diagnostic Letter Interpreter
Ever felt lost reading a medical report? This agent is ready to assist! Medical reports are commonly written for healthcare professionals and often leave patients with little clarity about their condition. The AI Coach has the ability to process medical records, summarize and explain them in easy-to-understand language.

The AI Coach is now ready to become an integral part of the Healactively mobile app. With its personalized support and smart features, it’s designed to help users take charge of their back health journey and make lasting improvements.

ImageInsight: AI-Driven Image Management Solution for Architecture

Students: Siriwat Suwattanapreeda and Luofei Pan

In the architectural industry, visual data is essential for project documentation, presentations, and planning. However, as image collections grow, traditional folder-based storage systems struggle to meet the demand for efficient image management. To address these challenges, Siriwat and Luofei developed ImageInsight for nts Ingenieurgesellschaft mbH. This AI-powered solution leverages smart search capabilities to enable efficient image retrieval and management, greatly enhancing accessibility, accuracy, and data security.

AI Image Processor for Automated Categorization and Privacy Compliance

At the core of ImageInsight is a powerful AI image processor designed to automate the generation of image tags and descriptions for fast retrieval, while also ensuring data security. This processor performs two main functions:

• Object Detection and Material Recognition: The processor analyzes each image to identify objects, materials, colors, and contextual details, generating detailed metadata that supports diverse search queries, such as “find an image with a park bench” or “show images containing glass material.” 

• Sensitive Information Removal: To ensure data privacy, the processor automatically detects and masks sensitive information, such as faces or license plates, ensuring that the image database complies with GDPR and other privacy regulations. 

 

Scalable Image Database: ImageInsight is equipped with a robust and scalable database to organize and manage nts Ingenieurgesellschaft’s growing image collection. 

User-Friendly, Multi-Language Website with Advanced Search Capabilities: ImageInsight’s website offers a user-friendly design and powerful multi-language search functionality. The platform supports search queries in French, German, and English, allowing users to enter keywords in multiple languages to retrieve relevant images. With this search functionality, users can quickly locate the images* they need and proceed with upload, download, and management operations.
 
ImageInsight provides nts Ingenieurgesellschaft mbH with an exceptional AI-driven image management solution, enabling efficient handling of visual assets through smart search, automated categorization, and privacy protection. With its scalable database, advanced multi-language search capabilities, and user-friendly interface, ImageInsight meets the unique needs of the architectural industry, setting a new standard for efficient and compliant image management.

*All images with people and cars in this blog post are AI-generated.
 

Automatic Detection of Solar Panel Disconnections

Students: Kittiboon Taopittayathorn, Yvonne Creter, Kristjan Mark

Fluence Energy AG is a global market leader in energy storage products and services, and cloud-based software for renewables and storage assets. Their department in Zurich specializes in providing data intelligence services for renewable energy worldwide. 

Solar energy production can be significantly affected by string disconnections and long-term degradation of photovoltaic (PV) panels. These issues often go unnoticed, leading to decreased efficiency and increased maintenance costs for solar energy providers. 

That’s where Kittiboon, Yvonne, and Kristjan took on this challenge. They worked on creating an innovative solution that integrates data from the solar park’s monitoring systems to detect faults and degradation in real time. They used a combination of rule-based labeling strategies, machine learning models, and deep learning to develop a system that identifies string disconnections on the Input, Combiner box, and Inverter levels (see picture).

 

The system processes inputs like electric power, current, and voltage from the solar panels, along with environmental data such as temperature and sunlight intensity. By training the models on over three years' worth of data, the team was able to build accurate predictive models with an impressive 93% accuracy for detecting disconnections at the input level and 82% accuracy for combiner box-level faults.



At the heart of this solution are three key models:

• Input-Level Model: Detects partial and full disconnections in the input strings of the solar panels.

• Combiner Box Model: Identifies issues in the combiner boxes that connect multiple solar panel strings.

• Inverter Model: Tracks the performance of inverters, ensuring they are operating optimally.

The project doesn’t just stop at detecting faults; it also provides operators with a user-friendly web dashboard that displays live data and machine learning predictions. Operators can receive real-time error messages and insights, which help them quickly identify which part of the system needs attention, reducing downtime and ensuring optimal energy production.



With the support of Constructor Academy's data science graduates, Fluence is setting a new standard in renewable energy monitoring, and we're excited to see how this innovative solution continues to evolve.
 

Conclusion

As we wrap up this remarkable journey with Data Science Final Projects Group #27, our heartfelt gratitude goes to the companies that provided invaluable projects for our students. Your collaboration has enriched their learning experience and contributed to innovative solutions for real-world challenges.

To the students who joined us in September and committed themselves fully to completing the course and their final projects, we commend your outstanding efforts. Your dedication, skill, and passion for data science have truly stood out. We wish you all the best in your future endeavors — may you continue to push boundaries, innovate, and make a meaningful impact wherever your careers take you.

For those inspired by these stories and eager to embark on their own data science journey, we’re excited to announce our upcoming programs. Learn more about our program and discover how you can join the next cohort of data science innovators at Constructor Academy.