By Daniel Ampuero Anca, Senior Software Engineer
A few weeks ago, our organization exemplified its culture of innovation and collaboration by hosting an electrifying idea-thon that brought together the creative minds of employees across departments. Several dynamic teams took on the challenge to ideate and develop solutions for real-world problems ranging from customer-facing challenges to internal knowledge management. The result? Two winning projects that tied for first place—each pushing the boundaries of innovation.
The idea-thon Experience
An idea-thon is an innovative collaborative event that breaks down traditional barriers in innovation competitions by offering three dynamic categories: Shark Tank, Flow’s Studio, and Hackathon. This unique format democratizes participation by welcoming individuals from diverse backgrounds, ensuring that technical prowess is not a prerequisite for entry. In the Shark Tank category, participants pitch entrepreneurial concepts and business models, challenging their strategic thinking and presentation skills. Flow’s Studio invites creative minds to explore the flexibility of our platform’s workflow capabilities to solve complex problems . The Hackathon track remains available for technical participants who want to develop working prototypes or software solutions. By providing these varied pathways, the idea-thon encourages cross-disciplinary collaboration, allowing designers, entrepreneurs, programmers, and visionaries to showcase their talents, exchange ideas, and potentially transform innovative concepts into tangible projects. This inclusive approach not only broadens participation but also fosters a rich, interdisciplinary environment where creativity and problem-solving skills are the primary currencies of competition.
In its third edition, the final presentations showcased a spectrum of ideas—some aimed at enhancing customer experiences, while others focused on improving internal workflows and processes. The competition was fierce, but two projects stood out for their innovation and potential impact.
First Winning Project: Hyperscience Everywhere (Model As A Service – MaaS)
The first winning team identified a common hurdle in the adoption of machine learning models: the need to purchase and maintain a complete platform, which can be costly and resource-intensive. To address this, the team developed a platform-agnostic solution allowing businesses to access and integrate ML models via APIs.
This platform acts as a bridge, enabling organizations to leverage cutting-edge ML models without having to commit to the overhead of the entire platform infrastructure. Key features include:
- Model Hosting: Companies can deploy models in a secure, scalable environment.
- API Access: Simple API endpoints allow users to send data and receive predictions seamlessly.
This project is impressive due its potential to democratize machine learning adoption, especially for small to mid-sized businesses that may not have the resources to build or manage complex ML infrastructures.
Second Winning Project: Extracting Text from Warped Pages
The second winning team focused on a practical challenge faced by many organizations: extracting text accurately from warped or distorted document images, such as those from curved book pages or unevenly scanned records. Their approach combined the power of existing tools with a structured analysis to evaluate and enhance outcomes.
The team began by assembling a specialized dataset of non-flat page images, capturing a variety of real-world distortions. They then leveraged an off-the-shelf page dewarping library to preprocess these images—or left them unprocessed in some cases to enable comparisons.
To assess the impact of dewarping on document usability, they wrote scripts and developed visualizations that provided detailed insights into:
- Segmentation Quality: How well the dewarped images preserved logical divisions of text and other elements.
- Reading Order: The accuracy of text flow, ensuring content was extracted in the intended sequence.
- Bounding Box Overlap: The degree of alignment between the dewarped output and ground-truth bounding boxes.
This meticulous analysis allowed the team to understand how preprocessing influenced downstream tasks like text extraction and layout interpretation. By identifying areas where the dewarping process excelled or fell short, they laid the groundwork for further refinement and optimization.
Looking Ahead
This idea-thon highlighted the incredible talent and creativity within our organization. Both winning projects will receive additional support for further development and possible implementation. The API-based model hosting platform and the improvements in image dewarping are expected to have a lasting impact on our offerings, enabling us to serve customers and stakeholders more effectively.
As we look ahead, the spirit of this idea-thon will continue to inspire our teams to innovate and push boundaries. Congratulations to all participants for their hard work and ingenuity—you’ve set the bar high for the next competition!
Stay tuned as we follow the journey of these winning projects and continue to celebrate the power of collaborative innovation.
