AgriPulse

The data to save the harvest already exists. AgriPulse is the bridge.

Comment

🏆 Built for DreamHacks 2026

🌍 The Harvest Ledger

By 2050, the world must feed ~10 billion people — on the same farmland, in a destabilizing climate. The data to do it falls from the sky every five days. We refuse to let it expire unread.

AgriPulse is real-time planetary monitoring for next-gen farming: open-source crop intelligence that fuses satellite imagery, weather, and soil into explainable, actionable insight for every field on Earth. Free forever. Self-hostable. MIT licensed.

Repository: https://github.com/robloxsagax-web/AgriPulse · Demo video: https://youtu.be/1RWzJamAJI0

🚨 The 2050 Problem

We don't get to shrug off this number. Feeding 10 billion people is the assignment of this generation — and the arithmetic says we're losing time.

The hard number What it really costs
🌡️ More volatile seasons Droughts, floods, heatwaves now arrive faster than farmers can react
💧 ~70% of global freshwater to agriculture Irrigation allocated largely without data — misallocated by design
🛰️ Every field imaged every 5 days, free Locked behind per-hectare analytics prices the smallholders — who produce one third of the world's food — can't pay
🌱 One third of soils degraded Yield and carbon bleed invisibly below our feet

The divide is not data; it's access. Farmers can't buy the analytics layer that would turn free orbital imagery into decisions. That is the entire AgriPulse thesis: put the orbital pipeline in everyone's hands, frankly explainable.

🌉 The Bridge

AgriPulse is open-source public infrastructure for crop intelligence — end-to-end, explainable to the last pixel.

Draw a field boundary on a map (or let our ML model detect it — PyTorch on Fields of The World), and the platform delivers automatically:

  1. ⏪ Backfills 24 months of vegetation history — NDVI, EVI, SAVI, NDWI from the public Sentinel-2 archive at 10 m
  2. 📡 Watches the field every 5 days and raises explained alerts — each anomaly ships with observation, magnitude, time window, and the weather + soil drivers behind it. No black boxes
  3. 🌱 Profiles soil to 200 cm (SoilGrids global / POLARIS US) and scores 68 crops for suitability with a transparent 4-pillar model
  4. 🌍 Quantifies sustainability — carbon sequestration potential, water balance, drought index, leaching/compaction/waterlogging risk

No API keys · No vendor lock-in · No per-acre fees · MIT License

🛰️ The Full Feature Dossier

Capability How it works Why it matters
🛰️ Satellite Intelligence 4 indices (NDVI, EVI, SAVI with configurable L, NDWI) at 10 m from Sentinel-2; auto 24-month backfill + weekly compute Trends appear on day one; anomalies aren't hindsight
🌦️ Weather Intelligence Daily history + 7-day forecast from Open-Meteo — temp, precip, ET₀, VPD, soil moisture/temp; agronomic derivations: GDD, water balance, drought index Explains the why, not just the what
🌱 Soil Intelligence SoilGrids (250 m global) / POLARIS (30 m US) — texture by depth, pH, OC, CEC, bulk density, AWC; 68-crop suitability (Soil 40 · Water 25 · Climate 20 · Stress 15) with limiting-factor explanations; sustainability: carbon sequestration, nutrient risk, sampling zones, soil×weather stress Transparent models you can defend in a boardroom
🤖 ML Boundary Detection PyTorch/TorchGeo (Fields of The World) — draw a rough area, review detections with confidence, accept in one click Eliminates the "magic" boundary-drawing step
🚨 Explained Alerts Configurable thresholds and drop-percentage rules; every alert enriched with weather + soil drivers No false panic. No unexplained noise
🗺️ Delivery MapLibre + PMTiles (no Mapbox token), colormaps, scouting markers with photos; ECharts percentile bands; read-only share links; OAuth → JWT, RBAC, audit logs, EN/ES i18n, themes Farmers, agronomists, buyers share one truth

🏛️ The Three Layers

Engineered as a 3-layer system — independently deployable, assembled to last.

┌───────────────────────────────────────────────────────────────┐
│  Layer C — DELIVERY SURFACES                 (Distribution)   │
│  Map UI · Reports · API · Webhooks · MCP · Mobile scouting    │
├───────────────────────────────────────────────────────────────┤
│  Layer B — INTELLIGENCE ENGINE                    (Moat)      │
│  Phenology · Anomaly detection · Stress signals · Yield       │
│  Risk models · Soil-derived insights · Explainability         │
├───────────────────────────────────────────────────────────────┤
│  Layer A — OBSERVATION INFRASTRUCTURE       (Data Gravity)    │
│  Satellite · Weather · Soil · Field boundaries · Sensors      │
└───────────────────────────────────────────────────────────────┘

Each layer is a module you can swap without breaking the others. That's the moat: explainability bounded by architecture, not by prompts.

📸 The Platform, Live

*Field health at a glance — vegetation index maps, 24-month trend charts, explained alerts with drivers, ML-detected boundaries.*

⚙️ The Machine Under the Hood

Stack Purpose
apps/web/ Next.js 14 · NextAuth (Google OAuth) · Tailwind · shadcn/ui · MapLibre · ECharts
services/api/ FastAPI · SQLAlchemy 2.0 (async) · Alembic · Celery
services/tiler/ TiTiler COG tile server (shared JWT auth)
infra PostgreSQL/PostGIS · Redis · MinIO · Docker Compose · Caddy (auto-SSL)
ml PyTorch · TorchGeo (Fields of The World)

✔ 7-service Docker Compose — one command; no cloud account; no paid keys. ✔ CI-enforced quality — ESLint + typecheck + build; Ruff + format check. ✔ Audit-logged RBAC — owner/admin/member/viewer with audit events on key actions.

Deep dive: ARCHITECTURE.md · DESIGN.md

📊 The Impact Ledger

Mapped to UN SDGs — measurable, not metaphorical:

Metric Mechanism Outcome
🌾 SDG 2 · Zero Hunger 68-crop suitability + yield risk Better crop-planning under climate change
🌡️ SDG 13 · Climate Action Per-field carbon sequestration estimation Track carbon stocks transparently
🌍 SDG 15 · Life on Land 24-month stress history + early anomaly alerts Detect decline in weeks, while restoration is possible
💧 Water stewardship Water balance, drought index, ET₀, VPD, leaching risk Allocate irrigation with data
⚖️ Equity $0/acre, self-hostable, MIT Satellite-grade intelligence as public infrastructure

🧗 The Hard Climbs

  • 🌐 Cross-source telemetry — STAC, weather, and soil come from different eternities; we normalized both time and geometry before the math runs.
  • 🤖 ML that doesn't own the field — boundary detection must suggest, let the user accept, and be honest about confidence. Detection without review is how bad fields get analyzed.
  • 📐 Soil scoring at scale — 68 crops demand a weighted model, tuned across texture, pH, AWC, and weather stress — enough structure to be plausible, simple enough to explain.
  • 🗺️ Tile economics — PMTiles + TiTiler beat SaaS tile costs by orders of magnitude; the tradeoff was freshness vs. compute, solved by scheduled re-inks.
  • 🔐 RBAC without friction — OAuth + JWT + org roles, auditable, with no surprise private endpoints.

📚 What We Learned

We arrived with production web habits and left with geospatial intuition: OG COG/PMTiles, self-hosted tile serving, correct geometry handling in PostGIS, distributed queues with Celery, and how to debug ML regressions honestly. The lesson we won't stop sharing: the explainability boundary is a design choice, and it's the cheapest moat we've seen.

🗺️ The Road Forward

On ROADMAP.md:

  • 🌾 Yield estimation & phenology tracking — monitoring → forecasting
  • 📱 Mobile scouting app — offline-first observations synced to the map
  • 🔌 Webhooks & MCP integrations — field intelligence in the tools farmers already use
  • 🤝 Community agronomy models — open, peer-reviewed crop and risk models

🌾 Why AgriPulse Demands Attention

💡 Unmatched Originality

AgriPulse breaks away from the noise of generic sustainability dashboards. It delivers a highly unique solution by fusing open-source precision agriculture directly into deeply explainable, per-field alerts.

🌍 Authentic Earth Stewardship

The platform achieves a total, uncompromising implementation of sustainable farming practices. It provides comprehensive, real-world utility by actively modeling water balances, tracking localized drought threats, measuring soil carbon retention, and running precision scoring analytics for 68 distinct crop varieties.

🚀 Production-Ready Completion

This is not a concept; it is an entirely completed product. The repository ships with a functional live demo, an instant one-command Docker deployment pipeline, strict CI quality control, multi-language internationalization, and professionally formatted, shareable field reports.

📚 Accessible Educational Framework

The project acts as an elite learning blueprint. It utilizes a modern, teachable technical architecture backed by immaculate, step-by-step documentation explicitly curated to mentor developers entering the agricultural tech ecosystem.

🎨 Human-Centric Design Principles

The interface prioritizes premium visual utility. It utilizes a highly structured, token-driven design system paired with the clinical readability of IBM Plex typography and fully compliant accessible color themes.

⚙️ Cutting-Edge Infrastructure

The software architecture relies entirely on modern, cloud-native pipelines. Data flows seamlessly from STAC to COG, TiTiler, PostGIS, and PMTiles, supercharged by PyTorch machine learning, asynchronous Celery task workers, and strict role-based access control security.

🔗 Deploy It Yourself

Prerequisites: Docker 24+ · one free Google OAuth client · two random secrets.

git clone https://github.com/robloxsagax-web/AgriPulse.git
cd AgriPulse
cp .env.example .env
docker compose -f docker-compose.yml -f docker-compose.dev.yml up --build

Then open http://localhost:3000, sign in, draw a field — the 24-month backfill begins. Production deployment (Caddy auto-SSL, Terraform/cloud-init): DEPLOYMENT.md.

🙏 Made on Open Data

Copernicus Sentinel-2 (ESA) · Element84 Earth Search · Open-Meteo · ISRIC SoilGrids · POLARIS · Fields of The World · OpenStreetMap & Protomaps

Built With

Share this project:

Updates

Submission history