SOOM Technologies LLC · SIU Carbondale

AgriLakehouse

A scalable, cloud-native data lakehouse unifying Genomic, Environment, Management, and Remote Sensing data modalities for real-time precision agriculture intelligence.

G — Genomic / SNP E — Environment M — Management RS — Remote Sensing AI-Powered by Claude
Platform Metrics
Modalities unified
4
G · E · M · RS
Records processed
2.4M
across all layers
Prediction accuracy
87.3%
R² on test set
Pipeline latency
4.2s
Bronze → Gold avg.
Lakehouse Architecture — Bronze · Silver · Gold · Serving
Bronze Layer
Raw Ingestion
GCS buckets · Kafka streams · SNP arrays · Sentinel-2 tiles · Weather APIs · Farm operation logs
Silver Layer
Cleaned & Joined
PySpark on Dataproc · ACID compliance · Schema validation · Cross-modal temporal alignment
Gold Layer
ML Feature Store
BigQuery · Delta Lake · G×E×M tensors · NDVI time series · Curated, ML-ready features
Serving Layer
AI & Analytics
GAT encoders · Temporal Transformer · Cloud Functions · Real-time yield prediction APIs
Data Pipeline Flow
Ingest
Raw → Bronze
→
Clean
PySpark / Silver
→
Feature
BigQuery / Gold
→
Predict
GAT + Transformer
→
Serve
Cloud Functions
AgriLakehouse © 2025 · SOOM Technologies LLC · Siraj, PhD Candidate, Mathematics (AI Track), SIU Carbondale · G×E×M Platform
G — Genomic / SNP Markers
Allele frequency by marker
SNP-1021
0.82
SNP-4530
0.67
SNP-7821
0.44
SNP-2209
0.91
SNP-6614
0.55
SNP-3301
0.73
Genomic embedding clusters (PCA)
E — Environment (Weather & Soil Sensors)
30-day temperature profile (°C)
Soil sensor readings
Moisture
62%
Nitrogen
48 ppm
Phosphorus
35 ppm
pH
7.0
EC (dS/m)
0.28
Potassium
52 ppm
M — Management (Planting & Irrigation)
Weekly irrigation events (mm)
Season management progress
Planting
Done
Irrigation
78%
Fertilizer
45%
Pesticide
20%
Thinning
60%
Harvest
—
RS — Remote Sensing (NDVI / Sentinel-2)
NDVI field heatmap (8×8 grid)
0.1 (bare)
0.9 (dense)

Hover a cell to see NDVI value

NDVI seasonal time series
G×E×M Yield Prediction Engine
Adjust the four modality inputs to compute a real-time yield prediction using the trilinear cross-modal attention model (GAT + Temporal Transformer).
Input parameters — all four modalities
Predicted yield
4.82 t/ha
Model: GAT + Temporal Transformer
Confidence: 87% · R² = 0.873
Feature fusion: trilinear cross-modal
Modality contribution to prediction
Scenario comparison
Low input
2.9 t/ha
minimal irrigation
Baseline
4.2 t/ha
standard practice
AI-optimized
5.6 t/ha
recommended config
Theoretical max
6.4 t/ha
ideal conditions
AI
AgriAI Assistant
Agricultural intelligence · G×E×M expert · Bayesian reasoning
Hello! I'm AgriAI, your precision agriculture assistant integrated with the AgriLakehouse platform. I can help you interpret G×E×M data, explain NDVI and remote sensing indices, discuss yield prediction models, or answer questions about the Bronze-Silver-Gold lakehouse architecture running on GCP. What would you like to explore?
AgriAI · ready
Try: