Signal over volume
A useful travel assistant must compress noisy social and local signals without losing important edge cases.
Evaluation
We evaluate models on risk detection, language ambiguity, and response latency.
Notes on building models that can reason over visual, textual, and situational signals.
A useful travel assistant must compress noisy social and local signals without losing important edge cases.
We evaluate models on risk detection, language ambiguity, and response latency.