ContextLens is a working demo of explainable behavioral prediction.
It streams raw telemetry for one mock telco subscriber from two fragmented sources —
an on-device SDK and cloud webhooks — maps each signal into a shared semantic space
(real embeddings, precomputed), and resolves them into a propensity score where
every percentage point is traceable to a signal.
Try the three scenarios: a clean session, a signal conflict resolved by
time decay, and a sparse session where the system refuses to predict
rather than guess. Then type your own signal in the feed panel and watch
it get embedded and scored live — or flip the time-decay toggle to see the counterfactual.
ContextLens
Explainable intent resolution — from raw fragmented signals to an auditable score