Menu
Close
Drug and human trafficking networks operate across country and state lines, but the agencies that hold the ground-truth case data each see only a fragment. Roughly 18,000 state, local, tribal, and territorial agencies record traffic stops, calls for service, and incident reports in their own systems, under their own retention rules. A rented box truck with a third-party lessee and a route deviation is an ordinary report in one county. The same lessee entity appearing two corridors away is the fact that changes its meaning, and nobody has both reports.
Federal systems such as N-DEx and NCIC hold the national picture but are query-oriented: an investigator must already know what to ask. Pooling local case data into a national repository has stalled repeatedly on legal, political, and security grounds. The gap is proactive pattern recognition, delivered without asking any agency to surrender its data and without any machine substituting its inference for the legal determinations the Constitution reserves to people.
ADDX TALLYMERE™ is a cloud-based federated machine-learning platform that brings the model to the data instead of the data to the model. Participating agencies train a shared trafficking-pattern model inside their own CJIS enclaves and send out only privacy-protected model updates, secured by secure aggregation and differential privacy. The national model returns to score local records in place. No case file leaves the agency, and no agency learns what another contributed.
A two-plane architecture keeps criminal justice information out of model weights entirely. A training plane learns from open, aggregate, synthetic, and federated sources; a CJIS-compliant inference enclave applies the trained model to identified local records at query time. Autonomous agents handle the labor between a pattern match and a decision: NIEM normalization, PII tokenization with agency-held keys, screening with feature attribution, corroboration against sources the agency already holds authority to query, deconfliction, and lead-package assembly with full provenance.
Every legal determination remains a human act. No agent establishes reasonable suspicion, enters a person into a 28 CFR Part 23 intelligence store, initiates surveillance, or applies for a warrant. The functions that protect rights run fully autonomously: retention and purge enforcement, expungement propagation, continuous bias and drift monitoring with authority to suspend a model, and immutable audit of every inference and action. The platform is built to the FBI CJIS Security Policy v6.0 on a FedRAMP High substrate, with agency-managed encryption keys and the CJIS Security Addendum executed by every ADDX engineer with access.
ADDX TALLYMERE™ is designed to compress the path from a routine field report to a corroborated, decision-ready lead package from weeks to hours. In the reference scenario, a corridor traffic stop reaches a trained analyst’s queue with corroboration and deconfliction complete in about forty minutes, and the analyst, not the machine, decides what happens next.
The approach gives agencies proactive pattern recognition trained on national signal without centralizing a single case file, implements civil-liberties compliance as executable architecture rather than policy documents, and produces provenance that turns defense discovery into a query rather than a scramble. Success is measured on lead precision, corroboration-to-conviction conversion, analyst hours saved per corroborated lead, disparate-impact ratios monitored against published thresholds, 100% Part 23 purge compliance, and zero uncorroborated model output in any charging document.
ADDX TALLYMERE™ is a design concept in pre-decisional development. Timings and metrics above are architecture design targets, not measured results from a fielded deployment.
