This section is The content in this section is supplied by GlobeNewswire for the purposes of distributing press releases on behalf of its clients. Postmedia has not reviewed the content. by GlobeNewswire Article content
New capability, being demonstrated at IBC2026, helps AI teams search, evaluate and assemble rights-cleared footage to match training requirements and accelerate model development
Sign In or Create an Account
or View more offersArticle content
SAINT JOHN, New Brunswick, Sept. 09, 2026 (GLOBE NEWSWIRE) — Versos AI, the video training data infrastructure layer for library owners and AI labs, today announced a new capability built with NVIDIA NeMo that lets AI teams describe the video dataset they need in plain language to direct the Versos platform to find, evaluate and assemble matching content into a structured training dataset. For AI companies, this shortens a process that can otherwise require days of manual searching, review, and dataset preparation. For owners of premium video libraries, it helps their content become more discoverable and usable to create stronger partnerships with AI companies.
Article contentWe apologize, but this video has failed to load.Try refreshing your browser, or
tap here to see other videos from our team.Article content
Story continues below
This advertisement has not loaded yet, but your article continues below.
Article content
Versos will demonstrate the new capability at IBC2026 in Amsterdam, September 11-14, 2026.
Article contentArticle content
Dataset requests for training AI models often combine highly specific content needs with technical and licensing requirements. A buyer may need footage depicting particular subjects, actions or environments while also meeting requirements for duration, resolution, format, language, production quality and permitted use. Fulfilling those requests can require teams to translate one detailed specification into multiple searches, manually inspect large numbers of results, compare individual assets against the requirements, and organize the selected footage for delivery.
Article content
Versos’ new capability allows a buyer to communicate requirements in a single natural language conversation. The system interprets the prompt, translates it into structured criteria and searches Versos’ scene- and frame-level video intelligence. It then grades candidate footage against the specification and assembles the matching assets into a dataset.
Article content
“AI teams should be able to search for the data they need to train new models the same way they would describe it to a colleague,” said Chris Keevill, co-founder and CEO of Versos AI. “The people building AI deserve to benefit from it just like everyone else, and our agentic interface makes that happen by streamlining the process for curating training data. NVIDIA technologies help us coordinate the work behind the experience, so the process is simpler for the user and more scalable for content owners.”
Article content
Story continues below
This advertisement has not loaded yet, but your article continues below.
Article content
Technology Behind the Workflow
These new agents carry out all the steps required to turn a buyer’s request into a usable collection of video data, while keeping rights and provenance central to the process. The workflow searches footage that has been prepared for controlled AI licensing and maintains the ownership, licensing and provenance information associated with the content. This shortens and clears a path for AI teams to obtain structured datasets with documented permissions. And for content owners, it creates a more scalable way to respond to highly specific requests without manually reviewing an entire library.
Article content
Versos built the agentic curation workflow on NVIDIA and LangChain technologies at three levels:
Article content
- The NVIDIA CUDA Toolkit accelerates the inference that runs inside the Versos Video Library Intelligence Platform, the analysis that turns raw footage into scene- and frame-level data.
- NVIDIA Nemotron Ultra runs the agents that interpret a buyer’s request and carry out discovery and curation of the video training data.
- LangChain provides the agent scaffolding layer that coordinates the steps between a plain-language request and the platform’s search, metadata, grading and dataset-assembly functions. The approach supports multi-step workflows, tool use and specialized-agent delegation while preserving the modularity needed to integrate models and infrastructure based on customer requirements.