satoru.bio · DigiShield Labs
悟る — to perceive, to awaken to understanding
Satoru uses large language models to extract and synthesise fragmented biological knowledge at scale — making decades of specialised research queryable and accessible for the first time. We begin where data is richest and most opaque: deep time.
Datasets
Systematic Microbiome Intelligence for Lost Ecosystems
The first comprehensive, queryable database of prehistoric oral microbiomes, aggregating ancient samples from published literature into a standardised, spatially-indexed corpus with authentication metadata. SMILE makes cross-study comparative research tractable for the first time — resolving the metadata fragmentation that has blocked systematic analysis of human-microbe co-evolution across deep time.
Antibiotic Resistance Gene Database · Deep Time
The first unified database of antibiotic resistance genes recovered from ancient biological material — establishing a pre-antibiotic baseline for AMR evolution research across deep time.
Approach
A tiered large language model pipeline processes scientific publications through sequential classification, sample reconciliation, taxonomic composition extraction, authentication scoring, and methodological metadata capture. Model selection is calibrated to task complexity; quality is validated against known ground-truth datasets.
Satoru is built and validated by a specialist in bioarchaeological data. The principal investigator holds a PhD in Archaeology specialising in stable isotope analysis and organic residue analysis — providing direct disciplinary authority to assess and correct extraction outputs.
All aggregated databases are freely accessible under CC-BY 4.0 licensing. Extraction methodology will be published for peer review and community replication. The infrastructure is designed to be extended, forked, and adapted across biological domains beyond the initial scope.
PostgreSQL 16 with PostGIS enables geographic and temporal queries across the corpus — supporting regional comparisons, site-level drill-downs, and spatiotemporal visualisation. Each sample is georeferenced at point level with SRID 4326 and linked to archival sequence accessions where available.
Academic partnerships, data access, and grant enquiries welcome.