Back to workflows
Creators & freelancersEnrich and routeFreelancer

Video footage clip finder

Search video by spoken words, visible objects, people, and scene changes.

What this workflow does

Search video by spoken words, visible objects, people, and scene changes. It resolves identity, enriches the case with approved evidence, and shows why the proposed route is appropriate before anything is assigned.

What you gain

Editors reach usable source clips much faster.

What the AI agent changes

How this worked before

Rule-based routing handled only clean, predefined fields and broke when the request was incomplete or written in natural language.

What the AI agent changes

The agent can interpret unstructured context, gather missing evidence, explain its recommendation, and pause when confidence is low.

Agent trigger

A new recording batch is ingested or an editor submits a clip query.

Inputs

  • Process context: video, transcript, speakers, objects, scenes, and project taxonomy
  • Approved policies, ownership, and exception rules

Agent flow

  1. 1

    Resolve the case identity and validate source freshness.

  2. 2

    Enrich the case with approved context and show every scoring signal.

  3. 3

    Prepare ranked source clips with exact timecodes and match reasons with confidence and exceptions.

  4. 4

    Create the approved assignment and notify its owner.

Human decisions

After step 3

The accountable process owner approves ranked source clips with exact timecodes and match reasons.

Outcome

  • Searchable clip shortlist
  • Evidence, exceptions, and audit trail

Guardrails

  • Do not route low-confidence identity matches automatically.
  • Use an allowlist of destinations and writable fields.

Risks and mitigations

A wrong identity match can send a case to the wrong owner.

Require deterministic identifiers and expose the routing rationale.

Sources and evidence