Building Real-Time Video Agents with VAST Data Engine

Most video AI demos stop at simple playback or offline analysis. Real-time video intelligence at scale requires ingesting streams, processing content, and retrieving meaningful insights instantly.

WHAT YOU'LL BUILD

A working real-time video agent powered by VAST DataEngine.

You'll implement a full pipeline: from ingesting video streams to generating summaries, detecting events, and retrieving relevant moments using embeddings.

By the end, you'll have a system you can run, tweak, and take back to your team, capable of processing video in real time, flagging key events, and integrating with downstream tools like Slack.

Your pipeline will:

Ingest video via event-driven triggers (S3 buckets)Generate LLM-powered video summariesDetect events from video streamsCreate video embeddings for semantic searchRetrieve relevant video segments using vector searchSend automated notifications for key events

KEY TOPICS

Event-driven architectures for video processingBuilding with VAST DataEngine for AI pipelinesLLM-based video summarisationVideo embeddings and vector searchDesigning scalable, real-time video pipelinesTranslating prototypes into production systems

AGENDA

4:00 PM — Doors Open: Welcome & Check-In: Arrival, registration, and opening remarks.

4:30 PM — Framing & Vision: What We’re Building and Why

4:45 PM — Live Demo: End-to-End Video Agent in Action

5:00 PM — Guided Build Part 1: Core DataEngine Foundations

(Connect to VAST lab, trigger functions, LLM integration)

6:00 PM — Break

6:10 PM — Guided Build Part 2: Production Features

(Video embeddings, vector queries, user-facing applications)

6:55 PM — Production Wrap-Up: Scaling to Real-World Systems

7:10 PM — Q&A & Next Steps

7:25 PM — Networking with Peers and the VAST Team

8:00 PM — Event Close

LEARNING OUTCOMES

By the end, you'll be able to:

Explain how VLM-powered video agents work in real-time production environmentsUse VAST DataEngine to build scalable pipelines for video ingestion and processingImplement an end-to-end workflow: ingest → process → summarise → embed → retrieveApply vector search to surface relevant insights from large-scale video dataDesign event-driven architectures for automating video intelligence systemsUnderstand how to take a prototype and extend it into a production-ready setupConfidently adapt and reuse the starter repo for real-world use cases

WHO SHOULD ATTEND

Intermediate to senior developers, ML/AI engineers, agent builders, and data engineers.

Industries: AI, Media & Entertainment, Financial Services

PREREQUISITES

Required:

LaptopComfortable coding in PythonFamiliarity with APIs and basic ML workflows

Helpful (not required): Experience with LLMs, embeddings, or event-driven systems

Setup: You'll connect to the VAST lab environment (no local setup required). Instructions sent 3-5 days before the workshop.

​Seats are limited: register now to secure your spot!