The Story-Loom Engine

The Problem with Linear Audio

In legacy applications, when a user enters a geofence, the app plays file.mp3. This rigid 1:1 mapping ignores the chaotic reality of driving.

A passenger might be passing a historic bridge at 110 km/h (requiring a 15-second summary) or they might be stuck in holiday traffic next to the same bridge (allowing for a 5-minute architectural deep dive). Linear audio cannot adapt to velocity, time of day, or driver engagement.

The DAG Architecture

Yonda abandons linear audio entirely. Instead, content creators author narrative fragments into a Directed Acyclic Graph (DAG) across four depths:

1. Layer 0 (Glance): Root node. 15–20 seconds. Factual and punchy for high-speed transit.
2. Layer 1 (Chronicle): The default documentary path (~2 minutes).
3. Layer 2 (Archive): Deep-dive leaf nodes (diaries, engineering specs).
4. Layer 3 (Cognition): Synthetic LLM nodes generated procedurally.

When the vehicle enters a polygon, the state machine initializes at the L0 root node. As the node nears completion, the engine evaluates a Curiosity Tension score (a blend of historical user retention, current velocity, and explicit inputs) to determine the next edge. If tension is high, it traverses vertically to L1 or L2. If low, it resolves the audio motif and exits to the ambient soundscape.

Layer 3: Generative ‘Shower Thoughts’

While L0 through L2 are strictly authored and verified by human cultural custodians, Layer 3 is synthetic. It mirrors the natural “mind wandering” of a long road trip.

We utilize a highly constrained Local RAG (Retrieval-Augmented Generation) pipeline:

1. The engine scrapes immediate OpenStreetMap tags (natural=scrub, historic=mine, elevation=800m).
2. It injects this telemetry into an LLM template designed for cosmic scale, engineering curiosities, or pop-culture satire.
3. The generated text is passed through local Text-to-Speech (TTS) and inserted as a dynamic edge in the graph.

Example: “We just passed that abandoned concrete grain silo. Given the 600mm reinforced walls, it has an incredibly high zombie-apocalypse survivability rating, assuming you could seal the lower auger chutes.”

Multi-Axis CarPlay Controls

To safely navigate this multi-dimensional graph at 100 km/h, we remap standard automotive media controls (AVRCP):

• Next Track (▶▶|): Horizontal Traversal. Skips the current graph entirely and jumps to the next physical waypoint.
• Double-Tap Next: Vertical Down. Artificially spikes the Curiosity Tension, forcing the engine deeper into L2 or L3.
• Previous Track (|◀◀): Vertical Up. Surfaces the graph back to L0 or aborts the story.

                   [ L0: THE GLANCE ]  <-- (Init on Geofence Entry)
                         |
          +--------------+--------------+
          |                             |
    (Low Tension)                 (High Tension)
          |                             |
   [ EXIT TO AMBIENCE ]         [ L1: CHRONICLE ]
                                        |
                                  +-----+-----+
                                  |           |
                            [ L2: ARCHIVE ]  [ L3: SYNTHETIC ]