Core Technology Stack — Designed Around Claude

Precision Intelligence for the Soil.

AgriSense is a concept-stage architecture for combining multimodal data processing with real-world grounding to test whether farmers can receive useful, safer first-line crop guidance.

Explore the Engine arrow_forward
Advanced Agricultural AI
sensors
Prototype View
Concept workflow preview

AI Architecture — Claude-first

AgriSense is designed as a Claude-first agricultural AI application. The MVP will test Claude image and text reasoning with vetted agricultural references and farmer feedback.

Planned Stack Anthropic Claude API (Vision + Reasoning) Claude API Monitoring Cloud Speech-to-Text v2 Google Maps Platform vetted source retrieval Cloud Run · Firebase
visibility

Claude Vision — Crop Analysis

Claude's image understanding will be evaluated on farm photos and reported symptoms. AgriSense plans to validate image-based crop issue detection across priority crops such as rice, maize, cassava, and sugarcane.

Claude
240+ Target Conditions
mic

Voice — Cloud Speech-to-Text v2

The MVP will test farmer voice input, starting with Vietnamese farmer language. Broader regional dialect support is a roadmap item after initial validation.

Voice-First Roadmap
description

source retrieval — Reduced Hallucination Risk

The planned response flow will retrieve selected agronomic references and provide them to Claude so draft responses can be checked.

Unified Claude Response Engine

The intended MVP flow sends crop photos, symptom text or voice transcripts, and selected context to Claude. The target output is structured JSON with possible issues, uncertainty flags, suggested next steps, estimated costs, and source citations where available.

hub
database
Agri-Databases
Linked to global seed & soil banks
trending_up
Market Data
Candidate commodity price sources
thermostat
Local Weather
Candidate weather context
satellite_alt
Satellite Context
Future verification option

Grounded in Reality, Not Hallucinations.

Unlike a generic chatbot, AgriSense is being designed to pair Claude with retrieval from vetted agronomic references so responses can be traceable to selected agronomic references. This approach still needs to be tested with farmers and agronomists.

  • check
    FAO & IRRI Open-Access Database Integration Candidate references include the International Rice Research Institute's disease library and FAO crop protection resources.
  • check
    Commodity Price Feed Roadmap Regional commodity data sources will be evaluated after the diagnostic MVP is working.
  • check
    Traceable Source Citations The planned output format includes source links so agronomists can review answer quality.

Data Privacy & Infrastructure Security

Claude is the planned primary AI engine, with supporting services for hosting and storage. The MVP will be designed with consent, regional deployment choices, access controls, and encryption in mind before any real farmer data is collected.

lock
Encryption by Design
Planned use of encryption, access control, and key management as the MVP matures.
shield_person
On-Device Inference Option
For later offline-first use cases, lightweight on-device models may be evaluated after the cloud workflow is validated.
gpp_good
Privacy Compliance Roadmap
Designed with consent-first data collection and regional privacy requirements in mind; formal compliance work will follow real deployment planning.
Secure Cloud Infrastructure
MVP
Target Architecture
Claude API · Cloud Run · Firebase
Technical Specifications

Target Benchmarks for MVP Validation

These are validation targets and cost assumptions, not measured production or pilot results.

speed

<10s

Response-Time Target

To be measured during MVP testing using crop image + symptom text/voice transcript + grounded context.

target

TBD

Diagnostic Accuracy

To be validated with agronomist review and documented test cases before any production claim.

bolt

Est.

Cost per Diagnosis

A working cost model will be checked against actual Claude API and hosting usage during MVP testing.

monitoring Detailed Validation Metrics — Planned

Metric Value Notes
Claude API Availability To measure Track during MVP usage in the Claude API and provider status
Voice Recognition Accuracy To measure Start with Vietnamese farmer language and expand only after validation
Offline Model Accuracy Future test AI Edge is a later offline inference direction, not part of the first MVP claim
Image Processing Low-latency Image pre-processing + upload for 12MP smartphone photo (varies by network conditions)
Concurrent Users Tested To test Load testing after MVP endpoint implementation
Cold Start Time To measure Cold starts and latency to be measured from real Cloud Run deployment
Monthly GCP Cost Estimate To be forecast from measured MVP usage, not from completed pilot data

Crop Disease Coverage Roadmap — 240+ Target Conditions

AgriSense's target taxonomy covers major diseases, pests, and nutritional deficiencies affecting Southeast Asia's staple crops. Candidate references include public datasets and open agronomic resources; no proprietary pilot image dataset is claimed yet.

grass

Rice

98 target conditions

  • Rice Blast (Magnaporthe oryzae)
  • Bacterial Leaf Blight
  • Sheath Blight (Rhizoctonia)
  • Brown Planthopper (BPH)
  • Tungro Virus Complex
  • Stem Borer
  • Nitrogen/Phosphorus Deficiency
  • + 91 more target conditions
spa

Maize

62 target conditions

  • Fall Armyworm (Spodoptera)
  • Northern Leaf Blight
  • Stalk Rot Complex
  • Downy Mildew
  • Maize Streak Virus
  • Ear Rot (Fusarium)
  • Zinc Deficiency
  • + 55 more target conditions
eco

Cassava

47 target conditions

  • Cassava Mosaic Disease
  • Bacterial Blight (Xanthomonas)
  • Cassava Mealybug
  • Brown Leaf Spot
  • Anthracnose
  • White Fly Infestation
  • Root Rot
  • + 40 more target conditions
yard

Sugarcane

33 target conditions

  • Red Rot (Colletotrichum)
  • Smut Disease
  • Top Borer
  • Rust (Puccinia)
  • Wilt Disease
  • Leaf Scald
  • Iron Chlorosis
  • + 26 more target conditions
update

Expansion Roadmap:

Coffee, rubber, and tropical fruit crops are future roadmap candidates. University or research collaboration will be pursued only after the initial MVP scope is validated.

System Architecture

Planned end-to-end data flow from farmer's smartphone to AI-assisted draft guidance and back — with Claude as the primary AI engine.

smartphone
Client Layer

Flutter Mobile App

PWA (Lite version)

Offline Capture / Sync

Firebase Auth

api
API Gateway

Cloud Run (Serverless)

Cloud Endpoints

Cloud Armor (DDoS)

Identity Platform

psychology
AI Engine

Claude

Claude Response Evaluation

Vetted Reference Retrieval

Speech-to-Text v2

storage
Data Storage

Firestore · Cloud Storage · BigQuery

map
External APIs

Google Maps · Weather API · Commodity Exchanges

monitoring
Observability

Cloud Monitoring · Cloud Trace · Error Reporting

info

Limitations & Known Constraints

What AgriSense cannot do yet — and what we're working to improve.

  • warning AgriSense does not replace professional agronomist consultation for complex multi-disease interactions.
  • warning Accuracy has not yet been validated; crop-by-crop performance must be measured during pilot testing.
  • warning Voice recognition must be tested in noisy field environments before support claims are made.
  • warning Offline capture is planned; on-device diagnosis and full treatment workflows require further validation.
  • warning Initial scope targets 4 crop categories; tropical fruits and tree crops remain future candidates.

Review the MVP Architecture

We are seeking technical feedback, cloud support, and early validation partners before building the first field-testable MVP.

Request Concept Walkthrough View Market Analysis