Post-Production Workflow Agent

Wrap faster. Guess less

Comment

Inspiration

Post-production is where films quietly stall. A shot sits waiting on VFX approval, sound can't start the mix until picture lock, color grading is blocked on a pending review — and often nobody notices until the delivery deadline is already at risk.

Today, that status lives in scattered spreadsheets, Slack threads, and inboxes. There's no single source of truth, and critically, no one proactively chasing down the blockers. A coordinator has to manually ping five different people just to answer a simple question: "why isn't this done yet?"

We wanted to build something that didn't just report on the chaos, but acted on it: an agent that behaves like a producer's most reliable assistant — always watching the pipeline, always asking "what's actually blocking final cut?", and always following up until it's resolved.

That's where the name came from too — in film, "that's a wrap" means the work is genuinely done. We wanted an agent that helps you get there faster, with less guessing along the way.

What it does

Wrap ("Wrap faster. Guess less.") is an agent that sits across the post-production pipeline — edit, VFX, sound, color, QC — and:

Ingests production status documents (shot lists, editor notes, review logs) and extracts structured state: shot ID, owner, status, dependencies Answers natural-language questions like "What's blocking final cut?" by tracing the critical path through the pipeline Automatically triggers a workflow via IBM watsonx Orchestrate when a blocker is detected — notifying the responsible team, logging a follow-up task, and escalating if it goes unresolved Gives producers a live dashboard instead of a guessing game

How we built it

Gemini + Google Cloud Agent Builder handle the reasoning core: parsing multimodal production documents and identifying blockers and critical-path dependencies IBM watsonx Orchestrate powers the action layer — when Gemini flags a blocker, Orchestrate runs the actual multi-step workflow (notify → track → escalate) Cloud Run hosts the backend and dashboard We designed the agent to be deterministic: given the same production state, it consistently identifies the same blockers and takes the same action path, so producers can trust the output isn't a coin flip

Challenges we ran into

Getting Gemini to reliably extract structured data from messy, inconsistent production documents (real production notes are rarely clean) Wiring a genuinely called-in-code integration with watsonx Orchestrate — not just referencing it, but making sure it triggers real, verifiable workflow actions Balancing "agent autonomy" with reliability — we wanted decisive action, not unpredictable behavior, which meant constraining the agent's decision logic more than we initially expected

Accomplishments that we're proud of

Building an agent that doesn't just chat — it takes real, traceable action through a live workflow integration Getting a deterministic, repeatable pipeline working end-to-end under hackathon time pressure Designing a demo that shows a genuine before/after: chaotic status → detected blocker → automatic resolution path

What we learned

Multi-agent orchestration is less about the AI being "smart" and more about designing clear handoffs between reasoning (Gemini) and execution (Orchestrate) Real production workflows have more edge cases and ambiguity than expected — good grounding data matters as much as the model itself A visible, deterministic action trail builds far more trust with users than a purely conversational agent would

What's next for Post-Production Workflow Agent

Support real-time ingestion from tools studios already use (e.g., Frame.io, ShotGrid) instead of static documents Expand beyond notifications to auto-drafting resolution options (e.g., suggesting an alternate reviewer if the primary owner is unavailable) Add analytics to help studios spot recurring bottleneck patterns across projects, not just react to individual ones

Built With

Share this project:

Updates