Real time meets real world: How AI is making decisions at scale (DAT202)
AI-Powered Real-Time Decision Making at Scale
Introduction
Evan, the Chief Product Officer of Aerospike, introduced the session on AI-powered real-time decision making at scale.
Aerospike is a cloud-agnostic, multimodal data platform used in data-intensive applications by companies like Rakuten, Snap, and Flipkart.
The goal is to build a data platform that allows more developers to solve more business problems using more data, with a focus on delivering predictable sub-millisecond latency and high throughput, while achieving market-leading compute and storage resource utilization.
Human vs. Machine Decision Making
Humans make about 35,000 decisions a day, with 227 decisions on what to eat.
Machines can make decisions much faster than humans, with the fastest neurological activity in humans occurring at about 20 milliseconds.
In contrast, Aerospike customers can process hundreds of records, feed them through multiple neural networks, and make real business decisions in a 20-millisecond window.
The history of the human-software relationship has evolved from rules-based software to machine learning, and now towards deeper models with advanced reasoning capabilities, enabling human-quality or post-human quality decisions without sacrificing speed and performance.
Real-Time Decision Making
Real-time decision making refers to operating within a fixed, known window, where the system becomes useless if it goes past a certain timeframe.
Examples include transactional fraud detection, product recommendations, and manufacturing supply chain.
The data sources for these decision-making systems include historical data, near real-time (sliding window) data, and real-time data.
Aerospike is engineered to handle all three data velocities, allowing customers to choose the appropriate storage strategy for their workloads.
Combining Classical ML and Generative AI
Classical ML models (e.g., logistic regression, random forest) can provide fast inference and scale economics, while generative AI models (e.g., large language models) bring capabilities like unstructured data processing and zero-shot generalization.
Companies are exploring ways to combine the strengths of these two model types, using the fast classical ML models within the real-time decision-making window, and leveraging the more powerful generative AI models for tasks like model grading, automated feature extraction, and result interpretation.
Autonomous Decision Improvement
The concept of "agents" and "agentic architectures" is seen as a compelling way to build systems that can combine models, talk to external systems, and reflect on their own learning and iteration.
The vision is for companies to own both the high-performance models used in real-time decision-making systems, as well as the agents or agentic software that can learn from various data sources and improve the automated decisions.
Aerospike provides the structured and unstructured data storage at the different data velocities to enable this autonomous decision improvement.
Customer Presentations
Tony Du from Dataminr shared how the company's AI platform provides real-time alerts on events, risks, and threats, using a combination of predictive and generative AI models to process data from various sources.
Victor from AppsFlyer discussed how the company uses Aerospike as a key-value database to store ad engagements and perform low-latency reads and writes at massive scale, while leveraging Kubernetes and Graviton-based instances to achieve cost-effectiveness.
These cookies are used to collect information about how you interact with this website and allow us to remember you. We use this information to improve and customize your browsing experience, as well as for analytics.
If you decline, your information won’t be tracked when you visit this website. A single cookie will be used in your browser to remember your preference.