Glass Biome · AI-native research infrastructure

Cultivation protocols from agricultural research, validated at the lab.

91 cultivars across 13 verticals. AI scans agricultural research papers, extracts cultivar × parameter × nutrient × outcome findings, and validates them at the USDA-ARS lab on campus. Each cycle adds to the database.

S.01 The Engine

Research Coverage

Live view of the evidence base: papers ingested, claims extracted, and connections discovered. Scans run against Europe PMC, Semantic Scholar, and OpenAlex with Unpaywall full-text enrichment.

Scan Extract Link Model Protocol

Research Verticals

13verticals
91crops
141nutrients
51parameters

Scan Frontier

Ontology

Core Entities

  • Crop — plant asset being analyzed.
  • Parameter — growth/handling variable (light, EC, pH, stress, harvest timing).
  • Nutrient — measured compound or antinutrient outcome.
  • Claim — extracted directional statement tied to a source span.

Depth Signals

  • Full Text — body text ingested (best depth).
  • Abstract Only — title + abstract only.
  • Title Only — title-level metadata only.

Direction + Confidence

  • Direction — positive / negative / nonlinear / undetermined.
  • Confidence — heuristic extraction confidence, not statistical significance.
  • Undetermined — entities found but no explicit increase/decrease signal.

Scan Lanes

  • Claim Yield — scanned lane produced claims.
  • Scanned, No Claim — scanned but no valid triple extracted.
  • Not Scanned — queued in plan, not run yet.

S.03 Pipeline Health

Pipeline health at a glance: papers ingested, claims extracted, links discovered, and extraction yield rates.

Pipeline Stages

Pipeline Overview

 

A.01 Where the Evidence Sits

Distribution of evidence across crops, parameters, and nutrients. Identifies where the strongest signals are and where coverage gaps remain.

Claims by Crop

Claims by Parameter

Claims by Nutrient

Crop × Nutrient Evidence Map

Cell intensity = claim count for each crop-nutrient pair. Hover for details.

Coverage Gap Matrix

Red cells = zero evidence for this crop-nutrient pair across the full ontology.

First-Order Aggregates

Aggregated crop × parameter × nutrient relationships. Each row summarizes all claims for a triple into a dominant direction and mean confidence — the basis for dose-response models and protocol recommendations.

Crop Parameter Nutrient Claims Direction Confidence

A.02 Extraction Quality

Extraction quality metrics. Direction indicates whether a parameter increases, decreases, or has nonlinear effects on a nutrient. Confidence reflects extraction certainty, not statistical significance.

Direction Distribution

Direction Signal Detail

Confidence Bands

Papers by Year

Reference Ingestion Depth

Nutrient Gap Status

A.05 Gaps the Engine is Filling

Nutrients ranked by research priority: human deficiency prevalence weighted against available evidence depth. High-priority gaps are where growing protocols could have the most impact on public health outcomes.

Nutrient Priority Index

Priority Overview

A.07 Complex Links

Pre-LLM structural links: dose-response fits, hidden nutrient synergies, and third-order triads. These are computed from co-occurrence and ontology rules — causal validation requires Opus extraction.

Dose-Response Models

Crop Parameter Nutrient Model Strength Direction N Mean Effect %

Hidden Synergies

Crop Nutrient Pair Relationship Bioavail. Score Deficiency Adj. Claims Note

Third-Order Triads

Crop Nutrient Triplet Anchor Triad Score Hypothesis

A.08 Mechanistic Predictions

First-principles pathway predictions: the inference engine maps (crop × parameter × nutrient) triples through KEGG pathways and signal mappings to predict nutrient directions without relying on empirical claims. Unconfirmed predictions are novel hypotheses warranting experimental validation.

Crop Parameter Nutrient Predicted Direction Confidence Category Pathways