FieldSense AI: Open-Source Multi-Agent AI That Takes Plant Health Investigation Into the Real World

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass*

What I Built

Most modern plant apps encourage the opposite of being present: you snap a photo, stare at a loading spinner, receive a confident but hallucinated diagnosis like "99% Late Blight", and spend the next 20 minutes doom-scrolling chemical remedies.
FieldSense AI flips this paradigm on its head.
FieldSense AI is an open-source, evidence-based field companion built for hikers, community gardeners, urban foragers, and botanical field observers. Its philosophy is simple: Step outside. Observe deeply. Let AI guide the discovery—not replace it. The screen is deliberately designed to be the shortest part of your outdoor experience.
Instead of dispensing premature pronouncements from an isolated photo, FieldSense AI functions like a seasoned field ecologist walking beside you:

1. It acknowledges uncertainty: It clearly delineates directly observed facts from plausible hypotheses.
2. It prescribes safe, hands-on field tests: Rather than guessing, it directs you to touch the soil 2 inches down, check stem rigidity, or inspect the underside of leaves for spores or webbing.

3. It operates anywhere on Earth: Thanks to local inference and an offline botanical evidence engine, it functions deep in state parks, rural trails, or dense forest canopies with zero internet or cell signal.

Demo

https://fieldsense-ai-ldw8uke4pfh6gyrpqiaowo.streamlit.app/

• Codebase & Architecture: Full repository with interactive React 19 interface, FastAPI backend, and LangGraph multi-agent pipeline is documented at github.com/NikhilRaman12/fieldsense-ai.
• Core Interface Views:
  • 🌿 Mission: Outdoor observation ethos and minimal-screen guidance.
  • 🔍 Field Observer: Real-time evidence collection, agent execution trace, and structured Field Investigation Reports.
  • 📖 Field Journal: Local observation history with geolocation and photo logs.
  • 📚 Field Guide: Verified botanical symptoms, differential diagnoses, and hands-on physical verification techniques.

  - ⚙️ Settings & Transparency: Active AI engine indicator (LangGraph + Ollama vs. Offline Botanical Engine), source transparency, and model configuration.

  Code

  NikhilRaman12 / fieldsense-ai

  An open-source multi-agent AI system for plant health investigation, combining evidence-grounded retrieval, LangGraph orchestration, and persistent investigation memory to support explainable agricultural insights.

  🌿 FieldSense AI

  Step outside. Observe deeply. Let AI guide the discovery—not replace it.

  A genuinely useful, fully open-source field-science application that encourages people to leave their screens, explore the outdoors, observe nature, and record discoveries. The actual outdoor activity is the main experience; the AI supports it rather than replace it.

  

  ---

  ⚡ Free 1-Click Deployment (Streamlit Community Cloud)

  You can deploy FieldSense AI for free in 30 seconds on Streamlit Community Cloud:

  1. Fork or open repository: https://github.com/NikhilRaman12/fieldsense-ai
  2. Go to share.streamlit.io and click New app.
  3. Select this repository, branch main, and main file path app.py.
  4. Click Deploy! — Your public demo link will be live at https://fieldsense-ai.streamlit.app!

  🎯 What FieldSense Does

  A user visits a garden, farm, park, or natural habitat and collects field observations. The AI coordinates 6 specialist agents to investigate possible explanations, identify missing evidence, and recommend the next physical outdoor…

  
  

  View on GitHub

  
  
  The repository is fully open-source under the MIT license:
• Repository URL: https://github.com/NikhilRaman12/fieldsense-ai
• Tech Stack:
  • Frontend: streamlit(demo), React 19, TypeScript, Lucide Icons, Vite
  • Multi-Agent Backend: Python 3.11+, LangGraph, FastAPI, Pydantic
  • Local Model Runtime: Ollama (Llama 3.2 / Qwen 2.5 / Mistral)

  - Offline Fallback: Deterministic Botanical Evidence Synthesis Graph

  How I Built It

  FieldSense AI is architected around an open-source AI core that respects the constraints of the outdoor environment.

  The Specialist Agents (LangGraph StateGraph)

  Instead of a monolithic prompt, FieldSense coordinates 6 focused specialist nodes sharing a unified InvestigationState:

• Vision Agent: Extracts macroscopic visual markers (interveinal chlorosis, marginal scorch, concentric ring lesions, powdery white residue).
• Context Agent: Ingests environmental telemetry—soil drainage, recent temperature extremes, watering cadence, and planting container type.
• Knowledge Agent: Matches observations against botanical field indices (citing verified horticultural extension bulletins and pathology databases).
• Hypothesis Agent: Formulates differential diagnoses rather than jumping to a single conclusion (e.g., contrasting Nutrient Lockout due to pH vs. Root Rot due to Overwatering).
• Field-Test Agent: Dynamically generates practical, physical tests the user can immediately perform with their hands, pocketknife, or hand lens.

6. Review & Synthesis Agent: Computes an evidence-to-uncertainty ratio and generates the structured investigation report.

Why Does Open Innovation Matter?

FieldSense AI is proof of why open AI is essential for applications connecting people to the physical world:

1. The Wilderness Has No Cell Towers

Proprietary cloud APIs are useless at mile 6 of a mountain trail or in a rural community garden with zero connectivity. By running open-weight models via Ollama or dropping cleanly to our deterministic offline botanical graph, FieldSense AI works identically on an airplane, in the backcountry, or during power outages.

2. Zero-Cost, Community-Owned Agriculture

Smallholder gardeners, permaculture students, and grassroots nature conservancies cannot afford per-API-call billing meters to monitor their crops. FieldSense AI is 100% free, runs on consumer hardware, and costs nothing to operate indefinitely.

3. Absolute Privacy & Geolocation Protection

Foraging spots, rare native plant coordinates, and private farm yields should never be uploaded to third-party corporate servers for model retraining. With local open AI, all images, coordinates, and journal entries remain strictly in local storage on the user's device.

4. Transparent, Auditable Botanical Logic

Commercial black-box vision classifiers notoriously hallucinate fungal diseases when confronted with normal seasonal senescing. With open-source LangGraph workflows, every node's reasoning, evidence attribution, and uncertainty thresholds are completely inspectable, forkable, and improvable by the scientific community.

Taking It Outside: A Real Field Test

During our field testing session in an urban garden plot, we observed severe chlorosis on Solanum lycopersicum (early-season tomato) lower leaves:

• What a standard closed app did: Declared "Early Blight (98% confidence)" from a single photograph, suggesting immediate chemical fungicide spraying.
• What FieldSense AI did:
  1. Correctly identified that yellowing was restricted to older leaves with no concentric rings (contradicting Early Blight).
  2. Noted recent heavy rainfall from habitat context.
  3. Prescribed a hands-on field test: "Dig 2–3 inches into the root zone with your fingers. Check if soil is waterlogged and smell the soil for sour anaerobic odors. If muddy, test soil nitrate or allow drying before applying any treatment."
  4. The user tested the soil: it was completely saturated clay. The issue was temporary nitrogen leeching due to overwatering, not blight! No fungicide needed; the tomato plants recovered in 5 days after drainage was improved. FieldSense AI kept the user engaged with the soil, not glued to a screen.

The goal isn't to make people spend more time talking to AI about nature. It's to help them spend more time observing nature for themselves.

FieldSense AI uses open-source intelligence to turn curiosity into observation, observation into evidence, and evidence into better questions about the living world