Coralfil Labs // Open Developer Cookbooks
Atlantis AgentSea Developer Cookbooks & Domain Recipes
The Atlantis AgentSea framework provides modular, reproducible developer recipes for building marine-aware multi-agent systems, training specialized ocean models, verifying Canadian coastal statutory compliance, and modeling marine physics. All code is public and open-source under Apache-2.0 on GitHub.
Public GitHub Repository
Access all runnable Python scripts, sample data, and schema specifications directly in the official GitHub repository:
Cookbook Domains
ML Training & Sensor Curation
PyTorch / QLoRAPipelines for ocean sensor telemetry sanitization, physical parameter bounds enforcement, license auditing, and parameter-efficient 4-bit LoRA model fine-tuning.
curate_marine_dataset.pyAutomated data scrubbing for ocean buoy telemetry, physical parameter clamping (TEOS-10 bounds), open data license verification (CC-BY-4.0, OGL Canada), and JSONL instruction tuning export.
train_marine_lora.py4-bit NormalFloat (NF4) double-quantized training pipeline targeting attention and MLP projections, reducing base memory requirements to 3.5 GB for edge and vessel deployments.
Canadian Marine Statutory Compliance
DFO & Transport CanadaAutomated regulatory compliance evaluation engine enforcing Canadian federal and provincial marine statutes across coastal engineering, acoustic telemetry, and restoration projects.
compliance_checker.pyEvaluates DFO Fisheries Act Section 35(2)(b) habitat authorizations, SARA Section 73 critical habitat spatial intersections, least-risk construction timing windows, and Navigable Waters minor works orders.
Marine Ecology & Acoustic Modeling
Graph & AcousticsEcological property graphs for coral holobiont microbiomes, multi-stressor environmental stability calculations (HSI), and underwater acoustic soundscape propagation.
holobiont_graph.pyModels coral animal hosts and photosynthetic symbionts using an openCypher relational graph with thermal tolerance offsets and nitrogen fixation fluxes.
species_env_correlator.pyCalculates the composite Holobiont Stability Index (HSI) from temperature, practical salinity, pH, aragonite saturation, and depth to predict bleaching and pathogen cascades.
acoustic_propagation.pyAinslie-McColm chemical absorption, spherical-to-cylindrical spreading loss, geoacoustic seafloor reflection across five sediment classes, and marine mammal hearing injury zones.
Marine Communication Zones & MapLibre
GeoJSON & RF PhysicsLocational maritime radio frequency and satellite communication modeling across nine modalities, with complete MapLibre GL JS vector styling and GeoJSON exports.
comm_zones_geojson.pyCalculates four-thirds earth radio horizons and two-ray seawater path losses across AIS, GNSS, GMDSS Sea Areas A1 to A4, satellite LEO and GEO broadband, coastal cellular, buoy mesh, and VHF.
Core Architectural Rules
- Reproducible Interfaces: Each recipe runs standalone or connects to standard tools including Relate openCypher graph databases, PyTorch, and MapLibre GL.
- Deterministic Physics Bounds: Sensor values and acoustic models enforce physical laws rather than unbounded neural predictions.
- Statutory Alignment: Coastal operations automatically cross-reference Canadian Fisheries Act and Species at Risk Act regulatory baselines.
- Edge-Ready Optimization: Quantization and model footprints are benchmarked for shipboard and off-grid field stations.