LeptonX standards proposals are developed in alignment with the world's leading standards bodies.
ITU-T AI/ML standards for trustworthy AI and data sovereignty in personal device contexts.
U.S. standards coordination for clinical AI processing pipelines and on-premise medical AI deployment.
Full alignment with NIST AI RMF and NIST 800-53 security controls for medical AI system deployment.
LeptonX is authoring standards proposals across five domains of local medical AI.
Standardized benchmark methodology for evaluating retrieval-augmented generation systems on clinical oncology query sets. Defines scoring rubrics, query taxonomy, and minimum accuracy thresholds.
ActiveCertification protocol for verifying that medical AI systems process protected health information exclusively on-device with zero external data transmission.
ProposedEvaluation criteria for voice-based clinical AI including pronunciation accuracy, response latency, empathetic framing, and clinical precision.
DraftFramework for patient-controlled storage and querying of VCF/FASTQ genomic files on personal hardware. Defines encryption standards, access controls, and portability requirements.
ProposedSpecification for UMLS/SNOMED-CT/RxNorm-based vocabulary compression enabling higher retrieval density while maintaining clinical semantic fidelity.
DraftTechnical specification for secure, read-only, time-bounded mobile AI containers for offline clinical use. Covers encryption, model quantization, lifecycle management.
DraftLeptonX publishes transparent benchmark data across all query dimensions. Methodology is open and reproducible.
v1 baseline vs v2 production pipeline (all fixes applied)
44-query suite · 6,602 document chunks
Foundation
Corpus Cleaning
Doc Compression
UMLS Encoding
Voice + Mobile
