Hey everyone,
A quick update from the Swenson Media team. Following 18 months of architecture work and support from the Google for Startups Cloud program, we officially shipped General Availability for FAST/OS on October 1.
The Problem We Solved
Physical film and television production is notoriously fragmented. Crews typically juggle 6 to 9 disconnected legacy desktop apps and spreadsheets. When a schedule shifts on set, manual phone tag creates a ripple effect of delayed call sheets, crew fatigue, and unforced union turnaround penalties (SAG-AFTRA Rule 1406 and DGA forced calls).
How We Built It With Google Tech
We designed FAST/OS as a unified, real-time operating canvas for active sets:
-
Deterministic Script Parsing via Vertex AI: We use Google Vertex AI to parse complex 120-page screenplay PDFs directly into relational datasets (1/8th-page breakdowns, scenes, cast DOOD, props, and locations) in under 45 seconds with zero manual data entry. Strict JSON schema enforcement eliminates model hallucination.
-
Auditory Copilot via Gemini Multimodal Live: On an active soundstage, producers and 1st ADs can’t stop to type into a laptop. We integrated Gemini Multimodal Live over bidirectional WebSockets to power Prodigi—delivering low-latency, hands-free spoken queries for schedule lookups, crew call times, and turnaround audits.
-
Local-First & <50ms Edge Sync: Soundstages and desert backlots frequently suffer from spotty connectivity. Our state architecture pairs Cloudflare Anycast edge workers with isolated SQLite tenant storage, guaranteeing local-first resilience with sub-50ms global synchronization when reconnected.
Current Milestone
-
GA live with Stripe self-serve billing at
swensonmedia.com -
Actively running across 20 indie and commercial beta productions
-
2026 SBDC Pitch Competition Winner
Huge thanks to the Google Developer community and startup mentors who helped us benchmark our WebSocket gateways and Vertex AI pipelines during early sprints.
If you’re building local-first architectures with Vertex AI or experimenting with multimodal voice agents in high-noise, real-world environments, we’d love to trade notes in the comments!