Contemplating plans for next.orcasound.net this year. Should I create a separate zulip channel and/or standup? Or start moving features to the production app live.orcasound.net?

Thesis: accessible interpretation and storytelling layer on marine audio and environmental data that allows for an integrated read of multi-dimensional conditions, in real time or in a research context

Potential partners / people I have chatted with: --

  • Maritime Blue Ventures / Port of Seattle
  • UW eScience Scientific Software Engineering Center (SSEC)
  • Microsoft (Hack for Good, Microsoft Garage)
  • MarineSitu (local startup doing related work with marine mammal detection for offshore energy clients)

Lower-effort goals (all front-end React/TS):

  • surfacing data filters in the UI (the filtering function already exists but I don’t have a consolidated UI component built)
  • building a chart that counts detections of each type of animal or sound (there is already front-end Regex that scans detection comments and pulls out keywords)

Higher effort goals (Python/FastAPI backend):

  • LLM generated summaries of current filtered data state, with RAG literature references -- user can ‘chat’ by changing filters
  • natural language chat interface that can run SQL queries and pull up sample data views -- user can change filters via chat
  • json snapshots at regular time intervals that give averages, stats, and combined data across listeners, AI audio processing models, and visual sightings (awaiting ambient-sound-analysis metrics from Valentina’s team to fill this out more)

Low-effort features that can be applied to Live already:

  • Candidates panel (right side of the map with filters and cards)
  • Map data (sightings, detections per node, audible radius)
  • Color theme using more of the brand palette
  • Expanded mobile navigation

High-effort / high-priority – this a blocker for all other goals:

  • React state management for filtering and displaying multiple data sources -- I am using Context API, but Paul needs to see a more robust solution using Zustand / Redux / etc

Features awaiting orcasound/ambient-sound-analysis repo work:

  • In-browser spectrogram generation from PSD parquet files
  • Candidate drawer with spectrogram (and “half height” to see underlying map view)
  • Live listening drawer with spectrogram