Filter up to 25 records against up to 10 yes/no criteria you supply. Each record comes back with keep / review / drop, a reason naming the criterion it failed, a 0-10 score and the per-criterion probabilities. Never drops a record it could not read or could not judge: those come back as review or keep, flagged. Built on a decision model rather than a chat model, so it is a flat per-call price with no token accounting on your side.

Score up to 25 company or contact records against an ideal customer profile: 0-10 fit score, qualified / review / disqualified verdict, reason label. Calibrated on 240 labelled leads: 0.879 accuracy, and no qualified lead ever disqualified or vice versa. Catches competitor name variants and subsidiaries of excluded accounts that string matching misses. Uncertainty widens the net: sparse records, unnamed contacts and low-confidence calls go to review, never to disqualified.