Purpose-built
health AI models
Four specialized models — each trained for a distinct domain of human health intelligence. Not general-purpose LLMs adapted for healthcare. Built with domain-specific alignment and evaluated internally before product use.
SFT + RLHF · Retrieval over a curated index of medical literature · Internally evaluated
AskMN v3.0
Herbal Intelligence & Human Health Model
AskMN v3 is Mother Nature AI's primary health intelligence model — trained using a large curated corpus of medical, scie…
Sylvia v2.1
Mental Health & Psychological Wellness Model
Sylvia 2.1 is a domain-specialized mental health AI model with deep training across clinical psychology, psychiatry, cog…
NutriGen v1.0
Nutrigenomics Intelligence Engine
NutriGen 1.0 is a multi-modal inference engine that fuses genetic variant data with dietary intake logs, micronutrient p…
Genlyy v1.3
Genomic Variant Intelligence Model
Genlyy 1.3 is a purpose-built genomic variant interpretation model that processes raw VCF files and SNP arrays from cons…
AskMN v3.0
Herbal Intelligence & Human Health Model
AskMN v3 is Mother Nature AI's primary health intelligence model — trained using a large curated corpus of medical, scientific, botanical and health literature from peer-reviewed research papers, clinical textbooks, and scientific studies. It is purpose-built to reason across integrative health, botanical medicine, clinical pharmacology, and human physiology with retrieval-augmented generation (RAG) over a curated medical literature index.
The model's training corpus includes the American Botanical Council HerbMed database, German Commission E monographs, European Medicines Agency botanical assessments, WHO monographs on selected medicinal plants, and the RTECS toxicology reference. AskMN v3 screens drug-herb interactions across 12,000+ compound pairs, plant constituent chemistry, bioavailability of herbal preparations, and dose-dependent safety thresholds — including identification of hepatotoxic, nephrotoxic, and cardiotoxic botanical risks.
Built on a transformer architecture with a 128K-token context window, AskMN v3 features a persistent health context engine that maintains coherent reasoning across lab results, supplement stacks, medication history, and wearable data trends simultaneously. Responses include evidence-level grading (strong, moderate, emerging) and contraindication flags.
Core Capabilities
- Comprehensive herbal medicine consultation — 3,000+ medicinal plants, constituents, and preparations
- Drug-herb interaction screening across 12,000+ compound pairs with mechanism-of-action explanations
- Plant toxicology assessment — hepatotoxicity, nephrotoxicity, cardiotoxicity, and overdose risk
- Supplement protocol design with evidence-level grading (strong / moderate / emerging / traditional)
- Lab result interpretation with trend analysis across longitudinal panels
- Symptom-aware guidance with botanical and integrative health context
- Wearable data fusion — HRV, SpO₂, sleep architecture, continuous glucose integration
- Traditional medicine systems reasoning — TCM, Ayurveda, Western herbalism, Indigenous plant medicine
curl -X POST https://api.askmn.ai/v1/chat \
-H "Authorization: Bearer $ASKMN_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "askmn-v3",
"messages": [
{
"role": "user",
"content": "My patient takes warfarin 5mg daily and wants to start St. John's Wort 300mg. Assess interaction risk and mechanism."
}
],
"options": {
"citation_level": "full",
"include_toxicology": true
}
}'Sylvia v2.1
Mental Health & Psychological Wellness Model
Sylvia 2.1 is a domain-specialized mental health AI model with deep training across clinical psychology, psychiatry, cognitive behavioral therapy (CBT), dialectical behavior therapy (DBT), trauma-informed care, and crisis intervention literature. Sylvia operates as a specialist inference layer within the Mother Nature AI+ platform, combining general health intelligence with rigorous mental health domain expertise.
The model is aligned to publicly available mental health classification and practice guidance, including ICD-11 and APA Clinical Practice Guidelines for major depressive disorder, GAD, PTSD, and bipolar disorder, SAMHSA evidence-based treatment protocols, and peer-reviewed psychotherapy outcomes research. Sylvia reasons across psychiatric comorbidities, medication-mood interactions, and psychosocial risk factors simultaneously.
Sylvia includes a dedicated crisis detection module trained to identify passive and active suicidal ideation, self-harm language, and acute psychiatric emergency signals. When triggered, Sylvia activates a structured safety protocol: immediate acknowledgment, risk-level stratification, and resource escalation to 988 Suicide & Crisis Lifeline, Crisis Text Line, and local emergency services.
Core Capabilities
- Evidence-based support for depression, anxiety, PTSD, OCD, bipolar disorder, and psychotic spectrum conditions
- CBT and DBT skill delivery — behavioral activation, thought records, distress tolerance, emotional regulation modules
- Crisis detection — suicidal ideation, self-harm, acute psychiatric emergency flagging with structured escalation
- Psychiatric medication awareness — SSRI/SNRI/MAOI side-effect profiles, mood-medication interaction screening
- Trauma-informed conversation framework — ACE score awareness, somatic response education, window of tolerance guidance
- Sleep and circadian rhythm coaching with evidence-based CBT-I (Cognitive Behavioral Therapy for Insomnia) protocols
- Burnout, stress, and resilience assessment using validated PHQ-9, GAD-7, and PCL-5 screening frameworks
- Psychoeducation delivery — attachment theory, nervous system regulation, polyvagal theory accessible explanations
curl -X POST https://api.askmn.ai/v1/specialists/sylvia \
-H "Authorization: Bearer $ASKMN_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "sylvia-2.1",
"messages": [
{
"role": "user",
"content": "I have been feeling hopeless for three weeks and have trouble getting out of bed."
}
],
"safety": {
"crisis_detection": true,
"escalation_enabled": true,
"safe_messaging_guidelines": "AFSP"
}
}'NutriGen v1.0
Nutrigenomics Intelligence Engine
NutriGen 1.0 is a multi-modal inference engine that fuses genetic variant data with dietary intake logs, micronutrient panels, and metabolic biomarkers to generate clinically-grounded personalized nutrition protocols. It is the first model in the Mother Nature AI family designed specifically for the intersection of genomics and nutritional biochemistry.
NutriGen analyzes 200+ nutrition-relevant genetic loci covering folate metabolism (MTHFR, MTR, MTRR), lipid and cardiovascular response to diet (APOE, LPL, PCSK9), fat mass and metabolic rate (FTO, MC4R), fatty acid metabolism and omega-3 response (FADS1/2), vitamin D receptor function (VDR), beta-carotene conversion (BCMO1), and methylation efficiency (COMT, AHCY). Each variant is contextualized against population-level effect sizes from GWASdb and aligned with NHANES dietary reference data.
Unlike generic nutrition models, NutriGen does not make population-average recommendations. Outputs are anchored to the specific variant constellation present in a user's genetic profile, cross-referenced against their current dietary logs and biomarker levels to identify actionable bottlenecks with the highest intervention leverage. Integration with Genlyy 1.3 provides full genomic context across all five health domains.
Core Capabilities
- Genotype-specific macronutrient ratio optimization based on FTO, MC4R, PPARG, and TCF7L2 variants
- Micronutrient absorption bottleneck identification — B12, folate, vitamin D, iron, zinc, magnesium
- Supplement stack generation calibrated to genetic variant gaps with interaction screening
- Food sensitivity risk flagging with genetic basis — gluten, lactose, caffeine, histamine, oxalate
- Omega-3 and omega-6 metabolism optimization based on FADS1/2 and ELOVL2 variants
- Methylation pathway support protocols for MTHFR C677T, A1298C and related variants
- Intermittent fasting and dietary pattern suitability scoring by genotype
- Integration with Genlyy 1.3 for whole-person genomic nutrition context
curl -X POST https://api.askmn.ai/v1/specialists/nutrigen \
-H "Authorization: Bearer $ASKMN_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "nutrigen-1.0",
"genetic_profile": {
"mthfr_677": "CT",
"apoe": "e3/e4",
"vdr_bsm1": "bb",
"fads1_rs174546": "CT"
},
"diet_log_7d": { "avg_omega3_g": 0.8, "folate_mcg": 180 },
"labs": { "vitamin_d_ng_ml": 22, "homocysteine": 14.2 }
}'Genlyy v1.3
Genomic Variant Intelligence Model
Genlyy 1.3 is a purpose-built genomic variant interpretation model that processes raw VCF files and SNP arrays from consumer and clinical DNA sequencing platforms to generate structured, provider-facing genomic reports. It is the most specialized model in the Mother Nature AI family, designed to bridge raw genetic data and actionable health intelligence.
Genlyy 1.3 analyzes 4,800+ clinically relevant variants across five major genomic health domains: cardiovascular risk (APOE, LPL, PCSK9, F5, F2), pharmacogenomics (CYP2D6, CYP2C19, CYP3A4/5, CYP1A2, SLCO1B1, DPYD, TPMT, UGT1A1), methylation and one-carbon metabolism (MTHFR, MTR, MTRR, AHCY, BHMT), neurotransmitter and cognitive function (COMT, MAO-A, MAO-B, SLC6A4, DRD2), and metabolic and body composition (FTO, MC4R, TCF7L2, PPARG, ADIPOQ). Variant allele frequencies are cross-referenced against gnomAD v4, 1000 Genomes Phase 3, and the UK Biobank.
All Genlyy outputs are structured as machine-readable JSON reports alongside provider-facing narrative summaries with inline citations. Reports include a medication metabolism profile cross-referenced against the FDA's pharmacogenomics table (230+ drug mappings), variant pathogenicity classifications aligned with ACMG/AMP 2015 guidelines, and a prioritized intervention list ranked by clinical actionability score — a composite metric derived from population prevalence, effect size, and clinical evidence level.
Core Capabilities
- Full VCF and raw SNP array ingestion — 23andMe, AncestryDNA, Illumina, whole genome compatible
- 4,800+ variant analysis across cardiovascular, pharmacogenomic, methylation, neurotransmitter, and metabolic systems
- CYP450 enzyme phenotype classification: poor, intermediate, normal, rapid, and ultra-rapid metabolizer
- ACMG/AMP 2015 variant pathogenicity classification with evidence tiering (P, LP, VUS, LB, B)
- FDA pharmacogenomics table cross-reference — 230+ drug-gene interaction mappings with clinical action levels
- Structured genomic interpretation: JSON output plus a provider-facing narrative with PMID citations
- Methylation pathway completeness scoring and one-carbon cycle bottleneck identification
- Population-stratified allele frequency contextualization via gnomAD v4 and UK Biobank
curl -X POST https://api.askmn.ai/v1/genomics/analyze \
-H "Authorization: Bearer $ASKMN_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "genlyy-1.3",
"input_format": "23andme_raw",
"analysis_modules": [
"cardiovascular", "pharmacogenomics",
"methylation", "neurotransmitter", "metabolic"
],
"output": {
"format": "structured_json",
"include_narrative": true,
"include_pgx_table": true,
"citation_level": "full",
"acmg_classification": true
}
}'Training & Alignment
How each model is built
Domain-specific alignment is central to each model. Releases are evaluated internally before they are used in our products.
Pre-training Corpus
PubMed and PubMed Central, botanical and clinical reference literature, botanical monographs, toxicology references, and publicly described clinical and genomic frameworks. Proprietary textbooks and licensed clinical databases are not represented as ingested training data.
Supervised Fine-Tuning (SFT)
Domain-specific instruction datasets reviewed by subject-matter contributors across herbal medicine, mental health, nutrition and genomics. Annotation is reviewed for consistency before use.
RLHF Alignment
Reinforcement learning from human feedback using preference data collected from subject-matter contributors, with regularisation to prevent capability regression.
Red-Teaming & Safety Eval
Adversarial prompt testing for harmful herbal combinations, crisis escalation failures, and pharmacogenomic misclassification. Ongoing internal safety evaluation.
Side by Side
Model comparison
| Feature | AskMNv3.0 | Sylviav2.1 | NutriGenv1.0 | Genlyyv1.3 |
|---|---|---|---|---|
| Context Window | 128K tokens | 128K tokens | 64K tokens | 64K + VCF |
| Latency P50 (internal eval) | < 1.2s | < 1.5s | < 2.1s | < 3.4s |
| Domain | Integrative Health | Mental Health | Nutrition + Genomics | Genomic Variants |
| Knowledge retrieval | ✓ | ✓ | ✓ | ✓ |
| Persistent memory | ✓ | ✓ | ✓ | — |
| Genomic input | — | — | SNP data | VCF + SNP arrays |
| Safety layer | Contraindications | Crisis protocol | Interactions | ACMG classification |
| HIPAA compliant | ✓ | ✓ | ✓ | ✓ |
| Provider output | ✓ | ✓ | ✓ | ✓ |
| Status | In production | In production | In production | Invite only |
Results shown are from Mother Nature AI internal evaluations. They have not been independently validated and should not be interpreted as current production performance. Scores are reported on held-out test sets not used during training or fine-tuning. Methodology is available on request.
Safety & Compliance
Built for clinical environments
HIPAA-Compliant Architecture
All model inference runs in isolated, encrypted compute environments. PHI handling follows HIPAA technical safeguard requirements, with audit logging of PHI access events.
Grounded Outputs with Citations
Mother Nature AI is designed to ground health information in retrievable medical literature and provide citations where available.
ACMG/AMP & ICD-11 Alignment
Genomic variant classifications follow ACMG/AMP 2015 guidelines. Mental health outputs align with ICD-11 and APA safe-messaging guidance.
No Training on User Data
Mother Nature AI does not use customer data or conversations to train, fine-tune, or improve its models. Conversations and personal health information stay private and remain yours.
API access is coming soon
Mother Nature AI is developing API access to its domain-specific health AI models. Public API access is not currently available. Register interest if you want to integrate these models later.