Hyper-local temperature, humidity, and soil-moisture checks designed for trained adults supporting schools and communities.
Open-source · Offline-first · Edge-ready
Climate intelligence where connectivity ends.
A transparent AI and IoT reference platform designed to help trained adults monitor climate risk in school gardens, community nutrition gardens, and vulnerable smallholder settings—without collecting personal or child data.
Current status: early reference implementation for supervised research and proposed pilots. Independent validation is required.
LOCAL ASSESSMENT
DEMO-001Latest synthetic reading39.0°CHeat stress reference threshold exceeded
Humidity82%
Soil moisture16%
Sync stateLocal
Combined reference risk90 / 100
Explainable reasons and next checks are available on-device.
Works without continuous internet
3non-identifying demo sites
360labelled synthetic records
7documented API routes
98.93%measured Python test coverage
00 — Reviewer brief
CLIMATE VENTURES ALIGNMENT · NO ENDORSEMENT CLAIMED
Directly aligned to published climate-tech needs. Deliberately honest about the evidence still required.
UNICEF’s published 2026 Climate Ventures themes include hyper-local heat and humidity alerts for schools, low-cost IoT monitoring, open-source frontier technology, and deployment in low-resource settings. FarmAssist maps to those technical needs through an offline environmental intelligence layer—without claiming validated health or nutrition outcomes.
Read the published Climate Ventures call ↗An edge-ready reference architecture for environmental sensors, local decisions, and delayed synchronization.
Functional software, reproducible tests, open licenses, human-readable configuration, and synthetic demonstration data.
Essential assessment stays local; connectivity improves coordination but does not gate the core workflow.
1 billion
children live in countries facing high climate and environmental risk.
466 million
Source: UNICEF Office of Innovation, Climate Ventures 2026 ↗children live in areas with at least twice as many extreme-heat days as in the 1960s.
01 — The challenge
Environmental signals matter. Connectivity should not decide who can act.
Climate stress can damage garden productivity, while intermittent networks can delay monitoring and response. Many tools also introduce unnecessary identity, location, or cloud dependencies.
FarmAssist demonstrates a different path: collect only essential environmental observations, process risk locally, show the logic behind every alert, and synchronize approved, non-identifying indicators when a connection returns.
02 — How it works
A complete local-to-public intelligence loop.
Four deliberately simple layers. No black-box claim, no identity layer, no permanent connection required.
OBS
Observe
Temperature, humidity, soil moisture, crop stage, and non-personal stress notes.
EDGE
Assess locally
Configurable thresholds generate risk levels, reasons, and adult next-checks on-device.
QUEUE
Queue safely
Approved observations wait locally, remain reviewable, and use UUIDs to avoid duplicate submission.
SYNC
Share aggregates
When connectivity returns, only privacy-reviewed environmental indicators move upstream.
Python 3.12FastAPISQLiteTypeScript PWAOpenAPIDockerMIT + CC BY 4.0
02B — Verifiable demo path
Five steps a technical reviewer can verify in minutes.
UNICEF’s published selection process verifies that a prototype exists, matches the proposal, and reflects the software and hardware work described.
Enter
Submit a synthetic environmental observation without personal data.
Assess
Run the configurable risk engine locally without continuous internet.
Explain
Inspect thresholds, reasons, confidence boundaries, and next checks.
Queue
Review an approved observation held safely on the device.
Sync
Reconnect and publish only non-identifying aggregate indicators.
Input contractStrict JSON Schema
Duplicate safetyUUID + HTTP 409
Model logicPublic JSON thresholds
Public outputAggregate only
Official selection process ↗03 — Explainability
REFERENCE ASSESSMENT · SYNTHETIC EXAMPLE
Every risk has a reason. Every reason leads to a check.
The current model is a deterministic threshold engine, not a predictive claim. Its JSON configuration is public, human-readable, versionable, and designed for local agronomic review.
SignalReadingRisk
Heat stressAt or above the 38°C reference threshold
39.0°CHighWater stressAt or below the 18% reference threshold
16.0%HighHumidity diseaseAbove the 75% reference threshold
82.0%ModerateReference model only · crop- and region-specific validation not yet completed
04 — Safeguarding by architecture
Designed to be useful without becoming invasive.
Data minimization is a product boundary, not a policy footnote. The open schema deliberately excludes identity, health, contact, household, and precise-location fields.
No child profiles
No child names, health records, biometrics, school-child identity, phone numbers, or personal farmer identity.
No precise locations
Public endpoints expose aggregate environmental indicators—not rows, notes, coordinates, or small-group breakdowns.
Adult human review
Alerts are uncertain prompts for trained adults, never autonomous instructions and never a responsibility placed on children.
Institutional safeguards
Consent, approval, access controls, retention rules, incident procedures, and accountable agronomic partners.
The platform is not a medical device, emergency-warning system, or substitute for qualified agronomic, safeguarding, nutrition, or public-health advice.
05 — Measurement
Built to measure progress without inventing it.
These are designed indicators for future supervised evaluation. No current outcome values are claimed.
01
Valid environmental observations
02
Observations processed offline
03
Synchronization success rate
04
Alert-generation latency
05
Active non-identifying sites
06
Sensor-data completeness
07
High-risk periods detected
08
Aggregate users trained
09
Future alert-interpretation rate
10
Connected-operation uptime
05B — Evidence ledger
Built, demonstrated, documented—or clearly marked as future work.
This ledger prevents prototype evidence from being confused with field validation or child-impact evidence.
Review questionEvidence availableStatus
Working prototype
API, risk engine, PWA, queue, synthetic generator
Built & testedOpen-source commitment
MIT software; CC BY 4.0 docs and synthetic data
Release-readyLow-resource deployment
Local processing, SQLite, installable PWA, Docker
DemonstratedChild-centred safety
No child profiles; aggregate-only publication boundary
DocumentedField validation
Supervised protocol, approvals, sensor and agronomic review
Not yet completedOutcome evidence
Pre-defined technical and future usability indicators
Future evaluation06 — Twelve-month pathway
From public alpha to evaluation-ready implementation toolkit.
A proposed sequence subject to funding, approvals, implementation partners, and evidence at each gate.
Months 1–2
Foundation
Architecture, safeguarding, licensing, schema, repository, baseline prototype.
Months 3–4
Offline product
Risk API, dashboard, local-first workflow, local-language framework.
Months 5–6
Edge & indicators
IoT ingestion, synchronization, public indicators, documentation.
Months 7–9
Pilot readiness
Approvals, training materials, usability and data-quality testing.
Months 10–11
Independent review
Model improvement, accessibility, interoperability, review preparation.
Month 12
Release pathway
Evaluation report, implementation toolkit, proposed stable release, scale plan.
06B — Execution foundation
JOITA BIOSEED AI PRIVATE LIMITED · INDIA
Built by a team already working across software, AI, IoT, agronomy, and field research.
The proposed child-climate module is new and unvalidated, but it is not starting without technical and domain foundations.
Offline-first FarmAssist test build and agronomic advisory workflows.
Programmer-led software development with AI and IoT experience.
Field-research workflows across three farmer locations covering approximately 25 acres in Haryana.
Laboratory and field-validation capacity for proposed supervised evaluation.
AgriTrust edge-hardware concepts and related industrial-design applications.
These capabilities support delivery readiness; they do not constitute validation of the UNICEF-specific module, model performance, child outcomes, or school deployment.
Proposed funding request
USD 100,000Funding would convert a transparent prototype into independently reviewable public infrastructure.
Planned work covers offline product maturity, reference IoT ingestion, accessibility and localization, safeguarding and pilot preparation, data-quality evaluation, independent-review readiness, documentation, and an implementation toolkit.
Validated architectureOffline productReference IoT adaptersSafeguarding packageSupervised-pilot toolkitEvaluation report
No UNICEF endorsement, selection, partnership, or funding is claimed.
07 — Why open
Public code makes the promise inspectable.
The repository includes the working API and PWA, risk configuration, schema, deterministic synthetic generator, tests, licenses, model and dataset cards, privacy threat model, safeguarding protocol, governance, and release history.
PUBLIC REPOSITORYv0.1.0-alpha
farmassist-child-climate-open
- ✓ MIT-licensed software
- ✓ CC BY 4.0 documentation and synthetic data
- ✓ Reproducible installation and testing
- ✓ Continuous integration and secret scanning
- ✓ Transparent limitations and change control
Open-source evidence liveThe complete v0.1.0-alpha prototype is public and independently inspectable.
Review the implementation, synthetic demonstration data, automated tests, governance documents, safeguarding controls, model limitations, and release history on GitHub.
JOITA BIOSEED AI PRIVATE LIMITED · INDIA