Portfolio

Developed a cloud-native, event-driven platform that scaled Canto's AI Visual Search to thousands of customers. This platform enabled large-scale processing of customer assets (images, videos) and real-time search across millions of these assets. This was achieved by implementing distributed inference for asset encoding and leveraging distributed vector databases for scalable, low-latency retrieval.

Building on my previous work on the Merlin Accelerated Intelligence Suite, the Canto AI Visual Search solution leveraged its machine learning models and search capabilities.

I also contributed to multiple AI research prototypes at Canto, most notably a hybrid search prototype that expanded AI Visual Search to include customer-specific asset metadata.

I played a key role in the development of the Merlin Accelerated Intelligence (AI) Suite, a comprehensive set of AI-powered tools that enhanced MerlinOne's digital asset management platform.

This suite powers the AI-based search for the AP Newsroom (try it out) and is also integral to Canto's AI Visual Search. Its success was a key factor in MerlinOne's acquisition by Canto in 2023.

My key contributions included the following:

• Visual Video Search: An high-performance, low-latency AI search engine for semantic video queries, allowing users to search for scenes by describing the visual content instead of metadata.
• Facial Recognition in Videos: A face recognition system enabling identity-based searches in videos.
• Model fine-tuning: Fine-tuning of AI models and search indices on customer-specific data for optimal performance.
• Inference pipeline: A highly scalable AWS inference pipeline to handle millions of images and video assets efficiently.
• System design: Advancing the system design and implementation of all AI services, encompassing a broad spectrum of areas such as asset processing, real-time and batch inference algorithms and search functionalities.

Designed and implemented an end-to-end ML platform covering the entire model lifecycle—from data handling to deployment.

The platform provided interactive data management, automated model training, versioning for both data and models, and comprehensive model governance. It also offered a centralized hub for monitoring model performance, tracking model versions and managing model deployments.

Overall, it optimized company-wide ML workflows, boosted efficiency, and enabled automated, reproducible model deployments.

In addition to the technical work, I also acted as technical and organizational lead of the machine learning team at cyan Security.

I enhanced cyan Security's child protection and content filtering products by training and integrating deep learning models.

My work involved training deep learning computer vision models to classify web page content based on visual features, improving the system's ability to filter inappropriate content and control access.

These enhancements to the cyan security suite were adopted by major telecom providers to safeguard their business and private customers.

Developed the content playout management platform for Red Bull TV's online streaming service, enabling reliable content delivery to millions of users worldwide.

Built on a reactive microservice architecture and distributed messaging infrastructure I was previously working on, the platform provided a highly scalable, fault-tolerant foundation for Red Bull TV's global streaming needs.

Designed and developed a distributed messaging infrastructure, to enable event-driven communication across all services in the client architecture. The system provided reliable asynchronous messaging, at-least-once message handling, and transactional support.

This infrastructure ensured reliable service communication, significantly enhancing system scalability and fault tolerance.

To make the messaging layer easier to use, I developed user-friendly client libraries to abstract the messaging layer's complexity.

Led the development of a reactive microservice framework used for all client-facing services, improving consistency and reliability across the organization. The framework was highly scalable and fault-tolerant, enabling the creation of reactive, high-performance microservices.

It provided common libraries for building reactive microservices, including application scaffolding, service discovery, monitoring and messaging, reducing redundancy and accelerating development across teams.

As the technical lead on multiple projects, I provided strategic oversight and guided the application development team in selecting tools and frameworks.

I also led the implementation of a RESTful web API for fund-related data and built a reporting application for private banking customers, seamlessly integrating it into the online banking platform.