Behind the scenes of Sim RaceCenter — how we build AI-powered broadcast production for sim racing.
Evergreen reference material that also powers the AI Director's context.
We ran a live GT3 race, exported the event stream, reconstructed the full race narrative from raw data, and discovered exactly where the AI Director would have failed to tell the story. This post walks through the methodology and the engineering improvements we've scoped as a result.
Broadcast domain knowledge and AI instructions for the Sim RaceCenter AI Director pipeline — view selection, sequence completion, camera resolution, race event types, and editorial guidelines.
Engineering narratives, post-mortems, and development updates.
A deep dive into how our Director App reads live iRacing telemetry using Node.js, Shared Memory, and Fast FFI — plus how you can build your own client.
How the Director app's iRacing extension was upgraded from streaming raw telemetry to publishing structured race events — and how this fundamentally changes what the AI Director knows about a race in progress.
After weeks of live testing the AI Director, we identified systematic gaps in template selection, timing, and content quality. This post documents what we found and designs five improvements — including Race Director's Notes and Session Summaries — to close the gap between technically correct sequences and compelling broadcast storytelling.
A post-mortem of a production incident where parse failures, rate limits, and connection timeouts cascaded to take the AI Director offline during a live broadcast.
How we designed a single wire format that carries every broadcast sequence — AI-generated, operator-built, or injected — from cloud to camera.
How the Director App processes broadcast sequences through a three-layer stack — from queue management and variable resolution in the Scheduler, through step dispatch in the Executor, to intent handlers in extensions.
How the Director App manages sessions and coordinates its subsystems through a mode-based finite state machine — from discovery through check-in to active broadcast direction.
A deep-dive into the two-phase Planner–Executor architecture that drives real-time broadcast decisions in Sim RaceCenter.
Domain knowledge for AI-powered broadcast direction of iRacing sim racing events — covering the platform, race procedures, camera systems, and professional directing principles.
A technical overview of the serverless backend that powers Sim RaceCenter — the AI Director pipeline, data model, and security architecture.
How the Director App reads iRacing shared memory via koffi FFI to extract session data, telemetry variables, camera groups, and live race state — powering the AI's understanding of what's happening on track.
How the Director App's extension architecture enables modular integration with iRacing, OBS, Discord, and YouTube through a two-tier registry, intent handlers, and a sandboxed utility process.
How the Director desktop app serves as the on-machine broadcast execution engine — managing extensions, executing sequences, and coordinating local integrations in real time.
An overview of Sim RaceCenter — what it does, who it's for, and how it transforms live sim races into professional-quality broadcasts.
