Conference Presentation, Keynote, Product Demonstration
Keynote by Christopher Savoie, Zapata AI CEO | RAISE Summit 2024 | Paris
- Zapata, founded in 2017 and spun out of a Harvard quantum science laboratory, operates as a public company utilizing tensor networks rather than traditional neural networks for AI training and inference.
- The company employs quantum math and tensor-based vector databases to compress vast operational and time-series data, enabling performance and accuracy at scale while reducing resource constraints.
- Zapata's strategic architecture favors an ensemble of specialized small language models (SLMs) and non-language models working in tandem over reliance on monolithic, billion-trillion-parameter LLMs.
- The firm explicitly targets edge deployment for generative AI, asserting that models are useless to enterprises if they cannot be executed locally due to latency, cost, or connectivity issues.
- Zapata partnered with F1/IndyCar racing to implement predictive analytics at the edge, processing live sensor data from cars traveling over 300 km/h in environments with intermittent cloud connectivity.
- During a recent race at St. Petersburg, an ensemble of five concurrent models accurately predicted a yellow flag probability increase at lap 26, preceding an accident at lap 27 and the flag deployment at lap 28.
- The company deployed a system of 60 models (30 ensembles of two) per car to predict real-time lap times and tire degradation by integrating streaming data with 20 years of historical records.
- Zapata utilizes generative AI to create "virtual sensors," synthesizing physical property data (e.g., accelerometers) with less than 1% error rate when physical sensors are impractical or prohibitively expensive to install.
- Edge inference is conducted on a military-grade server truck containing two A100 GPUs, specifically hardened to withstand high temperatures (exceeding 70°C) and vibration from the "damper factory" environment.
- Forward-looking applications of this time-series and edge-generation technology include biomanufacturing, advanced manufacturing scheduling (already reported with BMW), and financial trading strategy updates.
- Zapata recently announced a partnership with Sumitomo Mitsui Trust Bank to apply real-time time-series updates for trading decisions and to predict telecom network disruptions.