Our goal has been, and still is, to build the IAN "presentation layer". Dominic and I talked last week and are harmonizing on the expectations of this build; now, it is being coined as the Interplanetary Alert Network Lens (IAN-Lens).

Independently, we both explored the Tensor flow Textual Similarities algorithms. it took each one of us 7-10 minutes to analyze just 10 CAP records (I don't think I completed 20 records). Therefore, we are convinced that we should build our own algorithm for:

1. Clustering the alerts by geographic area and any other reliable CAP attributes
2. Apply the classification for that cluster of data (Bayesian for real-time classifier)
3. Offer a dashboard with the classified results

Assumption is that events of interest are almost always specific to a geographic location. This would reduce noise and latency. Moreover, it make better sense of the content to be densely clustered closer to the center of the event. We want to ready the algorithms to begin testing the hypothesis - "CAP is necessary and sufficient to serve as a basis classifier (or labeled training set per say).