Aporia launches root cause analysis tool for real-time production data analysis

Aporia, a machine learning (ML) observability platform, today announced the launch of a tool that aims to ease investigation of production data. The company asserts that its Production Investigation Room (Production IR) tool provides data scientists, ML engineers, and analysts with a “one-of-a-kind” unified monitoring platform that offers a digital environment for real-time data analysis, root cause investigation and deep insights.

Traditionally, investigating production data has been complex and time-consuming, hindered by limited collaboration and code changes.

Aporia claims that the new tool simplifies the process with a user-friendly and customizable interface reminiscent of a notebook. This should eliminate the need for extensive coding and help stakeholders derive valuable insights from their production data.

“Production IR provides centralized access for investigating AI/ML production data. [It] eliminates the challenges and pains of traditional methods, such as restricted data access, limited collaboration and the need for extensive code writing,” Liran Hason, cofounder and CEO of Aporia, told VentureBeat. “Through Aporia’s direct connection to the user’s database (DDC), it enables quick and efficient access to big data, simplifying the handling of large datasets.”

Hason emphasized that centralized visualizations of production data fosters collaboration and expedites root cause analysis (RCA). 

He argues that this approach improves ML model performance and enhances the efficiency and effectiveness of data exploration. The platform also empowers investigators to leave notes, report progress and alert others about specific issues, facilitating collaborative investigation.

According to Aporia, the new offering provides high customizability to cater to specific needs and can be easily configured to accommodate different datasets and requirements, enabling effortless visualization of investigations. 

Furthermore, Production IR automatically configures big data queries, alleviating the challenges associated with large-scale production models and data analysis. 

The company said the collaborative nature of the new tool promotes knowledge sharing among users. It enables comparison of analyses and facilitates sharing of insights within the Aporia platform.

“ML engineers and data scientists can leverage its capabilities to create interactive dashboards that can be shared and integrated with preferred tools such as Databricks, Snowflake and more,” added Aporia’s Hason. “[With] a unified view of data and insights, all team members can access the same information.”

Streamlining root cause analysis through unified data monitoring

Hason pointed out that traditional root cause analysis (RCA) relies on extensive coding, which consumes resources, causes delays, isolates insights and increases the potential for human error. Additionally, RCA is typically associated with high costs.

“Production IR overcomes these challenges by providing insights for improving models. [It] offers customization options, and provides an engaging experience for data scientists and engineers, fostering a collaborative investigation,” he explained. “This leads to accelerated mean time to resolution (MTTR) and simplifies the RCA process by improving response speed and agility while reducing the number of resources invested in tasks.”

With a wide range of analysis features, Production IR aims to streamline data investigation, encompassing segment analysis, data statistics, drift analysis, distribution analysis and incident response.

“Aporia’s segment analysis feature enables investigators to break their data into smaller, more manageable segments. This allows for a granular examination of specific subsets of data, which can help identify patterns, anomalies or correlations that may not be apparent when looking at the data as a whole,” said Hason. “Our platform’s new features empower investigators with analytical capabilities that enable them to conduct more efficient and effective investigations.”

Responsible and ethical AI, reliably and efficiently

Aporia claims that the tool’s incident response capability enhances AI products’ reliability and efficiency, enabling decision-makers to effectively address issues or threats. The company said that organizations can proactively tackle potential challenges by integrating incident response into AI practices and ensuring responsible and ethical AI deployment.

Furthermore, the tool incorporates an embedding projector, allowing users to visually represent unstructured data in 2D and 3D using UMAP dimension reduction.

“An embedding projector is a tool that helps users visualize and explore complex unstructured data, such as text or image data, in a lower-dimensional space, usually 2D or 3D visualizations,” said Hason. “It utilizes a dimension reduction technique called unified manifold approximation and projection (UMAP). This can be easily observed in the embedding projector visualization.”

Hason said the feature is significant for NLP, LLM and CV models, as it provides a comprehensive understanding of production data and drives improvements in ML models.

He explained that the embedding projector analyzes data points’ spatial arrangements, proximity and geometric relationships to uncover patterns within the data. These patterns expose underlying structures, trends or associations that may not be readily apparent in the original high-dimensional data.

“By leveraging an embedding projector with UMAP, users also gain a deeper understanding of their unstructured data, enabling tasks such as data analysis, model interpretation, feature engineering and hypothesis generation in the domains of NLP, LLM and CV,” Hason told VentureBeat. 

What’s next for Aporia? 

Hason said that Aporia aims to democratize and expedite the use of AI, enabling businesses to establish trust and ensure safe use. He pointed out that the consequences of AI errors can vary from mere inconvenience to potentially life-altering impacts.

“Imagine if the AI system in healthcare misdiagnoses a patient’s condition or if a financial prediction model fails to predict market trends accurately. The repercussions can be serious. It’s thus crucial to ensure AI systems are not just effective, but also reliable, understandable and trustworthy,” he said. 

Hason stated that Aporia has dedicated itself to assisting enterprises in achieving responsible AI through its ML observability platform. He emphasized that the platform enables transparency by offering clear insights into AI decision-making, fostering user trust and expediting the adoption of AI.

“At Aporia, our primary goal is to guarantee and enable responsible AI for every individual worldwide. We’re dedicated to building a platform that delivers an end-to-end solution for enterprises to handle their AI systems responsibly and effectively,” he said. “Our endeavor is more than just creating technology; it’s about establishing a safe and trusted environment for AI usage across all industries.”