Letter from our CTO 2024

Modena, December 2024.

The end of the year is the perfect time for reflection and building traditions. I'm glad to continue our conversation by sharing my second annual "Letter from the CTO," highlighting the incredible growth and achievements of Axyon AI's Tech Team.

After laying strong foundations in the past years, 2024 has been about scaling new heights. It's been a year of innovation, evolution, and delivering even greater value to our clients and partners. 

This is a curated selection of technological improvements made by our Tech Team in 2024:

1. IRIS ML Model Development v3.4 → v3.8

• What: We iteratively improved our ML Model Development pipeline from v3.4 (late 2023) to v3.8 (Nov 2024). Notable changes include (i) yearly ensemble re-optimization, based on predictions rather than metrics, (ii) combination of classification and regression models thanks to LTR, and (iii) new exploratory techniques to promote the discovery of nonlinear relationships. This was largely enabled by the parallel development of “Eolo”, an Apache Airflow-based component that streamlines and automates model development and retraining pipelines.
• Why it matters: Improved ML models produce enhanced predictive performance (therefore, alpha!) and robustness, boosting customer satisfaction, conversion rates, and revenue from performance-based deals. Eolo minimized human effort in development and retraining, reducing errors and improving efficiency.
  

2. Omakase/CODEX

• What: CODEX is a centralized data repository automatically built using Omakase, a new software component powered by modern data stack technologies like Snowflake and DBT. Omakase seamlessly integrates multiple data sources into a Single Version Of Truth (“SVOT”) database through a highly maintainable and scalable Data Engineering pattern.
• Why it matters: Simplified and standardized data integration, ensuring all downstream applications and services at Axyon AI have access to clean and trusted data. This significantly reduces development complexity, enhances overall system reliability, and provides a scalable paradigm for the incremental integration of additional data sources.
  

3. Talos

• What: Talos is a GUI-based software component that streamlines the dataset generation process with step-by-step guidance and integrated data quality assurance checks.
• Why it matters: Reduced dependency and toil on Data Engineering for dataset creation, allowing self-service for other Teams (Quant Research and Data Science). Furthermore, dataset generation is now significantly more streamlined, with tested and documented steps, reducing the risk of human errors and enabling faster project execution.

4. Horde

• What: Horde is a meta-project transforming IRIS production software into a fleet of Docker-containerised services running independently and seemingly communicating together. This multi-container pattern improves isolation and lifecycle management. In addition, when container configurations are changed by developers, Docker images are automatically built, tested, and pushed to a Docker registry, ready to be used in production.
• Why it matters: It accelerates the development cycle, significantly reducing dependencies between components, which can be independently upgraded. The streamlined container deployment reduces setup times and ensures robust, consistent live environments, paving the way for future DevOps improvements.
  

5. Yoda

• What: Yoda is an advanced Machine Learning model serving system capable of deploying heterogeneous ensembles, enabling better predictive performance through IRIS v3.8, and ensuring scalable and versioned deployments.
• Why it matters: Improved ML model ensembles allow for higher predictive accuracy, boosting alpha generation. Furthermore, Yoda has proven to be extremely scalable and reliable, producing more than 20 million predictions and 2 billion SHAP values with zero errors since its deployment in April 2024.

It's important to remember that all of the projects were team efforts and would not have been possible without the dedication, passion, and countless work hours of Team Leads and senior and Junior Developers in our Tech Team. I couldn't be prouder of our incredible technological journey. 

Together, we're thrilled to keep pushing boundaries, embracing new possibilities, and delivering even more outstanding achievements.

As we enter 2025, we remain committed to using Machine Learning to drive innovation and unlock new levels of success.

Here's to a new year filled with opportunities and alpha.

Best regards,
Jacopo Credi
Chief Technology Officer & Co-founder