SetWatch

Inspiration

Film and TV production plans contain dozens of facts, but only a few external assumptions can derail the day. Those assumptions may change after the call sheet is written and before the crew moves. SetWatch focuses live research on that moving edge.

What it does

SetWatch turns a production plan into falsifiable external assumptions, researches them through Parallel Search, and returns a bounded GO, VERIFY, or CHANGE operational brief.

Each material finding clearly separates:

• Evidence
• Inference
• Production consequence
• Recommended action
• Source trail

The demonstrated result identifies that a central London plan assumes no filming permit or parking dispensation is required. Live City of London evidence contradicts that assumption, producing a CHANGE decision before vehicles and crew are dispatched.

How we built it

Every live pre-flight begins with a mandatory Parallel Search call. Gemini, running through Google’s Agent Development Kit, evaluates the retrieved evidence and maps it to production consequences.

The application uses:

• Gemini on Vertex AI
• Google Agent Development Kit
• Parallel Search API
• Google Cloud Run
• FastAPI and Python
• Schema-validated structured outputs
• Runtime partner traces and evidence-integrity checks

Parallel is indispensable: without its current-web evidence, SetWatch can read the plan but cannot perform the live pre-flight check the product exists to provide.

Challenges

The main challenge was preventing incomplete search evidence from becoming false certainty. SetWatch treats an absence of contradictory results as insufficient proof and returns VERIFY whenever the available evidence cannot support a stronger conclusion.

We also made the partner integration consequential rather than decorative. A live decision cannot be issued unless Parallel Search succeeds first.

Accomplishments

SetWatch is a functioning production-oriented agent rather than a general chat interface. It provides a bounded decision vocabulary, live source trails, explicit Parallel runtime evidence, reliable failure behaviour, and clear separation between retrieved facts and operational reasoning.

What we learned

The useful unit of live research is not everything related to a production plan. It is the small set of external assumptions whose falsification would change an operational decision.

Parallel establishes the current evidence surface; Gemini converts that evidence into production consequence.

What’s next

Next steps include scheduled re-checks as the production date approaches and integrations with call-sheet and production-management systems, allowing changed external conditions to surface automatically.

Built With

• fastapi
• gemini
• google-adk
• google-cloud
• google-cloud-run
• parallel
• parallel-search
• vertex-ai