SensorPersona: An LLM-Empowered System for Continual Persona Extraction from Longitudinal Mobile Sensor Streams

SensorPersona addresses a valid gap in current LLM personalization: the reliance on self-disclosed chat history which is often incomplete or performative. By leveraging longitudinal mobile sensor data (accelerometers, GPS, app usage), it builds a more objective behavioral 'persona.' However, the project's defensibility is extremely low (Score 2) as it currently functions as a research artifact (0 stars, 6 forks, 33 days old) rather than a platform. The primary threat is 'Frontier Risk' from Apple and Google; these entities own the sensor stack and are already deploying 'Personal Context' features (e.g., Apple Intelligence, Gemini on Android) that integrate this exact data at the OS level. A third-party library has no moat against a system-level service that has native, privacy-preserved access to the raw sensor streams. While the academic approach of using LLMs to interpret sensor data into natural language personas is a clever 'novel combination,' it is an incremental step that frontier labs are likely already testing or shipping. Displacement is expected within 6 months as OS-level AI matures.