Three patents. One architecture.
Zero compromise.

120mm × 70mm × 25mm ruggedised housing, IP54 rated. 8MP camera module for capturing crop tissue, leaf surfaces, root systems, and soil conditions. GPS module for geo-tagging every observation. Connectivity via 4G LTE, 5G, Wi-Fi, and LoRaWAN. 12-hour battery life — a full day in the field.

Designed for farmers with limited literacy. Two buttons: Capture and Replay. That's it.

Compressed convolutional neural network running entirely on-device. Classifies 200+ crop diseases, 50 nutrient deficiencies, and 30 irrigation stress conditions. Diagnosis in under 3 seconds. Works completely offline — no internet connection required for core functionality.

Six dimensions of contextual intelligence:

• Crop Intelligence — what they grow, historical yields, disease patterns
• Soil Intelligence — composition, nutrient history, amendment records
• Yield Intelligence — seasonal performance, comparative benchmarks
• Micro-Climate Context — localised weather, altitude, humidity patterns
• Behavioural Intelligence — interaction patterns, learning pace, compliance
• Linguistic and Cultural Context — dialect, vocabulary, comprehension level

When the AI detects uncertainty, it escalates to a human advisor automatically. Three escalation triggers: confidence threshold breach (the AI isn't sure), repeated scan threshold (the farmer keeps asking about the same condition), and distress signal analysis (behavioural patterns indicating confusion or urgency).

The human advisor receives the farmer's full Intelligence Profile before the call connects. No cold starts. No re-explaining.

Every field observation generates a Geo-tagged Evidential Data Unit (GEDU) — an image linked to GPS coordinates, timestamp, and device identifier, sealed with a cryptographic hash. GEDUs are admissible as evidence in crop insurance assessment.

Aggregated across thousands of devices, GEDUs produce real-time geo-spatial maps of crop health, pest spread, and nutrient stress at village, district, and national level.

It is hardware-agnostic. It is domain-agnostic by design. It protects the data moat. It enables data licensing. A competitor can change the hardware. They can change the software. They cannot avoid the data architecture.

• D1: Domain Activity Record — Every advisory interaction logged with context, outcome, and temporal sequence.
• D2: User Knowledge & Competency Profile — Competency ratings derived from observed behaviour, not formal assessment.
• D3: Linguistic & Communication Profile — Local Domain Vocabulary Index, dialect, comprehension level, unit preferences.
• D4: Contextual Environment Record — Field, site, and location-level environment data.
• D5: Behavioural & Engagement Intelligence — Replay frequency, follow-up rate, advisory acceptance patterns.
• D6: Outcome & Efficacy Record — Domain outcomes linked back to advisory events.

No advisory event is treated as independent of prior history. A farmer who didn't understand advice last week receives the same topic recalibrated this week — without re-explaining anything. The system maintains complete continuity across every interaction.

The User Intelligence Profile architecture is domain-agnostic. The same six-dimension data structure that personalises agricultural advisory can personalise advisory in any knowledge-intensive field environment: industrial safety, vocational education, field service and maintenance, community health advisory.