Skip to content

Multi-Agent System for Media Asset Management

2025 – 2026 - ADK-based agentic workflow automation for video production at Forzasys

Overview

At Forzasys, video editors and content teams for sports clubs and broadcasters spend significant time on repetitive media tasks: finding the right clips, trimming them, exporting in the right formats, and publishing to social media. This project replaced much of that manual work with an ADK-based multi-agent system that automates media asset management and video-production workflows.

Agents built with the Agent Development Kit (ADK) are orchestrated over tool-calling, connected to the backend systems that power search, editing, export and publishing.

The Challenge

  • Fragmented workflow: producing a social-media clip touched several systems — asset search, editing, export, and each social platform
  • High volume, tight deadlines: sports content is only valuable while it's fresh; highlights need to go out minutes after the moment happens
  • Varied requests: "make a vertical clip of that goal with the celebration" doesn't map to a fixed pipeline — it needs reasoning about intent

Solution: Agents Orchestrated over Backend Tools

⚙️ Tech Overview

  • Multi-agent architecture: specialized ADK agents coordinated through tool-calling, each responsible for part of the workflow
  • Backend tool integrations: agents connected to tools for searching videos, simple video editing operations, exporting clips, and preparing content for publishing to social media
  • Stack: ADK, Python, Flask, tool-calling agents, media asset management, video processing, social-media automation

How It Works

  1. A user describes what they want in natural language (e.g. a clip, an export, a post)
  2. The orchestrating agent breaks the request into steps and delegates to the right tools — search the asset library, cut and export the clip, prepare the post
  3. Results come back for review, keeping humans in control of what goes public

Results

  • Reduced manual editing effort across the content pipeline — tasks that took an editor multiple steps across systems now run from a single natural-language request
  • Integrated in the production Forzasys platform used by sports clubs and broadcasters
  • The tool-based architecture made it straightforward to add new capabilities (new export formats, new platforms) without re-architecting

This project is a practical example of multi-agent systems delivering real workflow automation — LLM agents reasoning over a well-designed tool layer, embedded in a production media platform.