Jacaranda Scout: plan a walk through this week's purple streets, then put the phone away

Description and use

Jacaranda Scout automatically plans a walk through likely flowering jacaranda trees, selects the best day before rain, and provides pocket mode with spoken cues only at tree stops.

Problem

Users want to walk through jacaranda trees that are flowering this week but lack reliable data on which streets have blooming trees and when to avoid rain.

Outcome

On 11 October 2026, a 2.7 km loop through Parkview was planned with an 83% bloom chance and a best day of Monday to avoid 8 mm rain. Pocket mode displayed distance to next tree and spoke cues only at stops.

Workflow

• 1. User inputs intent (e.g., "easy 4 km walk from Zoo Lake with the dog").
• 2. IntentParser (local Ollama LLM) extracts activity, distance, start place.
• 3. DataHub gathers iNaturalist sightings, known tree locations, and Open-Meteo 7‑day forecast.
• 4. TabPFN sidecar predicts P(flowering today) for each known tree.
• 5. BloomScorer, HotspotClusterer, and LoopPlanner compute a loop using orienteering algorithms.
• 6. BestDayPicker selects a day before forecasted rain.
• 7. FieldGuideWriter (LLM) generates briefing and spoken cues.
• 8. App launches; in pocket mode, screen stays black and GPS geofences trigger spoken cues at each stop.
• 9. User walks the loop, receives cues only at trees, and can export GPX for watch/OsmAnd if desired.

Evidence

The app kept falling back to demo data.

Source

The model gives the trees on the route an 83% chance of flowering today.

Source

The loop came out short. I asked for 4 km and got 2.7 km.

Source

With TabPFN loaded, the first Ollama call on my 16 GB laptop failed to allocate its buffer.

Source