In continuation of its efforts to grow presence on hyperscaler infrastructure, last year, Striveworks announced delivery of its MLOps platform, Chariot, on Azure and Google Cloud. The integration follows a similar deal with AWS in 2022.
Originally released in 2020, Chariot is aimed at unstructured data processing applications like computer vision and natural language processing (NLP).
Data analytics processes of machine learning take up significant time and toil, leading to longer time to value. Chariot, a Kubernetes-based solution, is designed to enable data scientists to build, deploy and maintain machine learning models for demanding applications like energy exploration, military and intelligence, faster and more efficiently.
Direct competitors like the AWS SageMaker provide comparable MLOps functionalities but lack a cohesive user experience and deep integration with specialized data like geospatial information. Other players like Dataiku have only just began venturing into unstructured data.
Chariot gives Striveworks an early head start. The strategic focus on unstructured data may prove advantageous in distancing from competitors in the space that primarily target structured data.
For industries that heavily rely on geospatial data—such as energy exploration, defense and intelligence—Striveworks provides a modular, but unified approach. With it, data science teams can develop solutions for a range of environments and data types.
Chariot is optimized for complex data types – multimedia and text – that meets the demands of geospatial sector and applications requiring computer vision and NLP, says the company, angling it toward applications in mining, gas exploration and military intelligence, where satellite imagery, sensor data and multimedia inputs are more valuable than traditional structured datasets.
The company offers Sky Saber, a geospatial application built on top of Chariot, to process geospatial imagery with computer vision models.
At its core, Chariot aims to address the fundamental challenges in modern machine learning model development, said the team during a briefing with Gestalt IT, an arm of Tech Field Day.
Chariot is designed to manage models that must contend with adversarial elements, data drift and diverse data inputs.
Drift detection is essential in dynamic environments where the underlying data changes frequently, and engineers need to ensure that model outputs remain accurate and reliable over time. Chariot’s built-in drift detection feature is designed to proactively monitor changes and alert users when models’ performance degrades.
In unstructured computer vision data, where inputs can shift rapidly, this capability is not only relevant but necessary for mission-critical applications.
One of Chariot’s top selling points is end-to-end support for the MLOps lifecycle, said Striveworks. Data scientists are often drawn into peripheral setup and configuration works, that are a distraction from the bigger model development job. Within Chariot, datasets, training runs and models are managed under a structured interface. This also includes workspaces for high-code environments like Jupyter Notebooks and VS Code, that offer high-level customization while making it simpler to move from experimentation to production.
Additionally, Chariot’s model catalog and workspace setup streamline model management, allowing users to access a deep history of training runs and fine-tuned configurations.
Data scientists can avail pre-built, no-code training blueprints that integrate with PyTorch, while also building custom “Training Blueprints” tailored for specific needs. The balance of out-of-the-box functionality and customization makes it adaptive to unique operational requirements.
Performance monitoring and model evaluation are crucial to the efficacy of any MLOps platform, and Chariot includes a highly detailed set of performance metrics that enables “fine-grained” comparisons between models, offering valuable insights for model tuning.
As an example, Chariot can assess how different object detection models perform on satellite data in a particular region of the world during a specific season, said Striveworks.
The performance analysis is particularly relevant for applications where accuracy and consistency are mission-critical, such as in defense and intelligence sectors.
Chariot’s UI is tailored to manage vision-based models that data scientists and AI engineers can effectively work on, from image classification to object detection.
Striveworks currently supports large language models (LLMs), and plans to expand it in the future, a development that could open up additional possibilities for customers in the intelligence and public sectors, enabling applications that rely on language processing, document analysis, and more.
The platform is deployed on-prem, in air-gapped networks, in data centers and in the cloud which allows users to leverage edge computing – as well as specialized cloud infrastructure.
Head over to Striveworks.com for more information on the product.
