VectraFlow: An AI-Augmented Data-Flow System

About

VectraFlow is a new data-flow engine designed to seamlessly integrate modern ML models with an extended relational framework for unstructured and multi-modal data processing. VectraFlow extends the relational model by introducing advanced semantic operations (based on vectors, LLM prompts, and general ML models), offering support for diverse AI-driven applications. With its unified execution model, VectraFlow enables both real-time streaming and batch processing, making it ideal for use cases such as continuous and agentic prompts, real-time copyright infringement detection, and event detection in live video streams.

Products

• VectraFlow: An AI-Augmented Data-Flow System
  Shu Chen, Alexander Lee, Duo Lu, Deepti Raghavan, Malte Schwarzkopf, Uğur Çetintemel
  NEDB Day 2025
  [Slides] [Poster]
• VectraFlow: Integrating Vectors into Stream Processing
  Duo Lu, Siming Feng, Jonathan Zhou, Franco Solleza, Malte Schwarzkopf, Uğur Çetintemel
  CIDR 2025
  [Paper]

Demo

Team

Faculty: Uğur Çetintemel, Deepti Raghavan, Malte Schwarzkopf

PhD Students: Shu Chen, Alexander Lee, Duo Lu, Franco Solleza

Master's & Undergraduate Students: Justin Chan, Simeng Feng, Michael Fu, Nicholas Kim, Evan Li, Akshay Mehta, Weili Shi, Jonathan Zhou