FarmSim

🌱 Inspiration

Farmers face complex, high-stakes decisions with shrinking margins for error. Climate volatility, labor shortages, and rising input costs mean a single wrong call can end a season. We built FarmSim to move beyond inaccessible jargon. We wanted to take a farmer’s specific profile and answer: "What does my field look like in three months if I don't change my irrigation, vs. if I do?" Driven by a love for open source, we combined real-world data with custom 3D visualization to make risk tangible.

🚜 What It Does

FarmSim is a multi-agent system that bridges the gap between raw data and human intuition:

Persona-Based Analysis: Scrapes hyper-local news and weather signals to build a specific "Farmer Profile."
Live Intelligence: Pulls from Open-Meteo, NASA POWER, and USDA to feed a Monte Carlo engine (5,000 samples) that quantifies risk via P10/P50/P90 distributions.
The Prediction Engine: A custom-built Three.js animation library generates a side-by-side visual comparison.
Visual Proof: Farmers see a "Traditional Chaos" animation (predicted crop damage/stress) vs. a "FarmSim Advantage" animation (showing the impact of suggested mitigation).

🔧 How We Built It

Multi-Agent Architecture: Used Claude 3.5 Sonnet to orchestrate "Scenario Families" and ElevenLabs to provide a humane, voice-cloned advisor experience.
Animation Engine: Developed a proprietary Three.js-based rendering layer to turn abstract simulation data into 3D time-lapses of field health.
Deterministic Core: A Python-based Signal Service handles the heavy lifting—integrating live API calls and running NumPy-powered risk math.

🏆 Accomplishments That We're Proud Of

Custom Animation Library: We didn't just use charts; we built a dedicated 3D engine to visualize agricultural "what-if" scenarios.
End-to-End Pipeline: Successfully connected real-time news scraping and Monte Carlo math to a voice-enabled UI in a single request.
Open Source Commitment: Built entirely on open-source APIs and models, ensuring the logic is transparent and reproducible for the global ag-tech community.

📚 What We Learned

Visualization is Clarity: A farmer seeing a 3D shriveled crop is more impactful than a spreadsheet.
Deterministic vs. Generative: Multi-agent systems thrive when LLMs handle the "humane" synthesis while Python handles the rigorous math.
Profile Matters: Moving from generic weather to a specific "Farmer Profile" (including local labor and market signals) makes the AI an actual partner, not just a tool.

🚀 What's Next for FarmSim

Advanced Growth Models: Replacing heuristics with PCSE/WOFOST peer-reviewed crop growth science.
Agentic Review Loops: Allowing the supervisor to automatically re-run simulations if confidence intervals are too wide.
Community Library: Releasing our Three.js Ag-Animation Library as an open-source tool for other developers to visualize soil and crop health.