Arandu Technical Specification

Current Specification

This overview is intended for investors and strategic partners. It describes capabilities, architecture in functional terms, and governance posture—without disclosing proprietary algorithms, prompt internals, or implementation recipes.

1) System Purpose and Scope

Arandu is a WhatsApp-first agricultural intelligence platform, engineered in Switzerland for reliability and clarity. The product translates Earth observation and environmental signals into concise, grower-actionable guidance—delivered where farmers already work: mobile messaging.

• Channel: WhatsApp as the primary interface.
• Interaction model: Short natural-language input and structured menu choices—suited to low-literacy and intermittent connectivity contexts.
• Outputs: Prioritized text advisories; optional field visualizations where they add scouting value without burdening data plans.

2) Runtime and Deployment Architecture

• Application tier: A high-concurrency API framework with strict input validation at the edge.
• Compute: Serverless container orchestration—stateless instances that scale with demand.
• Heavy work: An asynchronous task queue decouples user-facing latency from geospatial and AI workloads; controlled fallback paths exist when queue delivery is unavailable.

2.1 Architecture at a Glance (Functional Stack)

3) Geospatial and Environmental Data Stack

• Managed geospatial compute executes analysis close to the data, reducing transfer and latency.
• Optical Earth observation informs vegetation and surface condition where atmospherically usable.
• Radar-capable sensing supports continuity when optical quality is limited.
• Weather and terrain context are fused for water, erosion, and operational planning narratives.

Aggregation and statistics are computed server-side; the product avoids chatty client-driven fetch patterns that would harm users on constrained networks.

3.1 Agronomic Signal Pipeline (Value-Focused)

Arandu’s differentiation is not a single index—it is the end-to-end fusion, calibration, and interpretation of signals into grower language. The pipeline produces compact, field-level summaries and optional sub-field views; exact weighting, thresholds, and fusion recipes are proprietary.

• Canopy vigor & biomass trajectory: Supports early stress awareness and seasonal benchmarking against the field’s own history—not a single snapshot in isolation.
• Nitrogen / chlorophyll stress (dense canopy): Targets late-season limitations of coarse greenness metrics where agronomic nuance matters.
• Water status & moisture stress: Relates canopy water signals to practical irrigation and dry-spell awareness.
• Erosion & land-loss risk: Combines terrain exposure and environmental drivers into a tractable threat lens for stewardship.
• Continuity under poor optical conditions: When optical quality drops, complementary sensing preserves a vegetation signal rather than returning silence—transparently flagged when applicable.

4) Analytical Modules (Public Capability Surface)

• Crop health and vigor.
• Nitrogen / chlorophyll-oriented stress proxies.
• Water status and moisture stress.
• Standing water and inundation-oriented signals.
• Soil erosion threat.
• Yield outlook and weather-informed field operations.
• Directed scouting visualizations where they sharpen on-ground inspection.

4.1 High-Level Flow: Report Generation

• The grower expresses intent (e.g., health or water focus).
• The platform ensures the request is legitimate, unique, and safe to process at scale.
• Analysis is orchestrated end-to-end for the active field.
• Field-relevant signals are derived and summarized for interpretation.
• A state-of-the-art LLM ensemble renders guidance within a fixed, validated response shape.
• Text guidance is delivered first; optional visuals follow when they add value.

flowchart LR
    user[Grower on WhatsApp] --> intent[Express analysis intent]
    intent --> trust[Ensure request integrity and uniqueness]
    trust --> orchestrate[Orchestrate field analysis]
    orchestrate --> signals[Derive field-level agronomic signals]
    signals --> render[Structured advisory rendering]
    render --> deliver[Deliver text guidance]
    deliver --> enrich[Optional field visualization]
                

4.2 High-Level Flow: Field Registration

• The grower anchors the field with a location and context.
• The platform establishes a trusted session and clarifies what is needed next.
• The field extent is captured through an assisted boundary workflow.
• The field identity is recorded authoritatively for subsequent analyses.
• Confirmation closes the loop; the field becomes eligible for monitoring and reports.

flowchart LR
    grower[Grower] --> anchor[Anchor field location]
    anchor --> session[Establish trusted session]
    session --> extent[Define field extent]
    extent --> record[Record field identity]
    record --> confirm[Confirm registration]
    confirm --> enable[Enable analysis and monitoring]
                

5) AI Interpretation Layer

• A state-of-the-art LLM ensemble interprets precomputed agronomic signals—not raw pixel streams in the chat path.
• Outputs are constrained to validated structured schemas, limiting unbounded narrative drift.
• Copy is tuned for mobile readability and operational clarity.
• Image understanding is invoked only when the user explicitly supplies imagery.
• Conversation context is bounded to balance quality, cost, and latency.

6) Outlier and Change Detection (High-Level)

Arandu emphasizes signal quality over novelty: separating seasonal noise from patterns that merit a grower’s attention.

• Historical context: Present-season signals are judged against multi-year baselines in aligned seasonal windows.
• Sub-field focus: Localized divergence from whole-field behavior supports targeted scouting.
• Trajectory awareness: Short-horizon trends and inflection points complement static thresholds.
• Operational intent: Alerts are designed to be explainable and actionable in the field—not academic dashboards.

7) Data Governance and Storage Characteristics

Arandu applies a Swiss-precision mindset to data handling: minimal surface area, durable records where needed, and disciplined growth.

• Authoritative conversation and field state reside in a distributed document persistence tier.
• Ephemeral process memory is an acceleration layer only—not a system of record.
• Stored artifacts are sized and bounded to protect performance and cost predictability.
• Personally identifiable information is minimized in logs; access patterns follow least-privilege discipline.

8) Reliability and Observability

• Continuous operational monitoring supports uptime and latency expectations.
• Structured telemetry underpins triage without publishing exploitable error fingerprints.
• Elastic compute and asynchronous work queues sustain availability during demand spikes and variable network quality.
• Release discipline includes pre- and post-deployment verification in production.

9) Known Constraints and Non-Goals

• Arandu is decision support—not a substitute for certified agronomists or statutory compliance advice.
• Remote sensing remains subject to atmosphere, revisit, and seasonality constraints.
• Recommendations are probabilistic interpretations of environmental signals, not guaranteed outcomes.
• This document deliberately omits implementation detail that would aid reverse engineering or competitive cloning.

10) Integration Surface (High-Level)

• Certified messaging channels for WhatsApp ingress and egress.
• Hyperscale cloud primitives for compute, queuing, and durable storage.
• Commercial and public Earth-observation and weather data ecosystems.
• Enterprise-managed AI inference for language and (where applicable) vision.

11) Data Security, Trust and Legal Notice

Data Security & Trust Statement
Arandu Technologies GmbH

At Arandu, we treat agricultural data with the same precision as the crops it represents. The platform is built on enterprise-grade cloud infrastructure so insights remain accessible while custody stays disciplined.

11.1 How We Protect Your Information

• Secure infrastructure: Stateless application tiers on leading cloud providers; durable state in managed, encrypted-at-rest persistence.
• Privacy controls: Personally identifiable information is masked in internal telemetry and processed only as needed to operate the messaging channel.
• Channel integrity: All provider integrations use industry-standard encrypted transport (e.g., TLS).
• No automated prescriptions: Arandu supplies decision-support signals; the grower retains final operational authority.

11.2 Intellectual Property and Legal Notice

The Arandu system architecture—including its proprietary fusion of geospatial analytics, robust statistical treatment, and LLM-based agronomic interpretation—is the exclusive property of Arandu Technologies GmbH.

© 2026 Arandu Technologies GmbH. All rights reserved.

Reproduction, distribution, or reverse engineering of the Arandu technical framework, logic flows, or proprietary prompt structures without express written consent is strictly prohibited.

12) Versioning and Change Management

This page reflects the externally shareable technical posture of Arandu. Material changes to architecture, analytical surface, or governance are reflected here in step with product releases.