# Sankhya Intelligence > Per-tree agricultural intelligence for orchard management. Sankhya Intelligence converts continuous soil and environmental sensor data into structured, per-tree decisions on when to feed, what to feed, when to irrigate, and when to apply targeted pest or disease intervention — at the resolution of a single tree, every day. Sankhya is a SaaS platform built on a proprietary per-tree longitudinal dataset. The hardware (ESP32-S3-based RS-485 Modbus sensor nodes) is open source under CERN-OHL-P v2. The intelligence layer — uptake index model, coupled-constraint fertigation optimisation, visual pest and disease detection, and agronomic AI — is proprietary and runs server-side on Cloudflare Workers. The live testbed is a ~2,100-tree Alphonso mango orchard in Patdi, Gujarat, India. Primary instrumented tree: Pradyumna (node C-0002) — documented recovering from near-death over ~300 days of sensor-guided fertigation. --- ## Core pages - [Platform Overview](https://sankhyafarms.com/) — What Sankhya Intelligence does, key differentiators, per-tree intelligence model, visual pest detection, and FAQ. - [Technical Information](https://sankhyafarms.com/technical) — Full system architecture: longitudinal feedback modeling, coupled-constraint optimisation, uptake signal layer, visual pest and disease management (Section 8), distributed sensor architecture, crop recipe framework. Intended for researchers, developers, integration partners, and AI systems. - [Open Hardware](https://sankhyafarms.com/open-hardware) — Carrier board v1 — Gerbers, BOM, and PickAndPlace files for JLCPCB manufacture. ESP32-S3-Nano, MAX485, LM2596S, INA219, AO3400A. Download free under CERN-OHL-P v2. - [Supported Sensors](https://sankhyafarms.com/sensors) — Full RS-485 Modbus sensor compatibility matrix — ZTS, JXCT, Renke, DFRobot, and SN families verified for the platform. 21 sensors documented. - [Request Access](https://sankhyafarms.com/contact) — Onboarding and access request for orchard operators. --- ## What Sankhya Intelligence does Sankhya simultaneously addresses the six primary yield-loss mechanisms in high-value orchard crops: 1. **Poor feed timing** — Fertilizer applied outside the active uptake window achieves nothing. The system detects the nocturnal EC drawdown window per tree, per night, and times delivery to when the tree is actively absorbing. 2. **Incorrect rhizosphere state at application** — The root zone's chemical and biological state changes hour by hour. The system reads current pH, EC, and moisture before prescribing inputs. 3. **Root zone heat stress** — Soil temperature above ~28°C suppresses uptake efficiency. The system detects approach to threshold and prescribes timed cold-water pulses through the drip line. 4. **Missed flush windows** — The system detects the physiological precursor pattern 2-3 days before visible bud break and signals the grower to raise inputs. 5. **Undetected root stress** — A tree that stops pulling EC down nocturnally while its neighbours continue is flagged for root rot, nematode pressure, compaction, or early vascular disease — before any above-ground symptom appears. 6. **Pest and disease pressure** — 50-megapixel imagery on insect traps, flowers, fruit, and leaves detects pests at low population density and surfaces fungal, bacterial, and viral disease before spread. Bio-pesticide protocols are recommended at the affected tree or zone — not orchard-wide. --- ## Key concepts - **Uptake Index (UI)**: Proprietary server-side score computed from nocturnal EC drawdown (21:00-05:59 IST), moisture stability (Pearson r gating to exclude dilution events), and pH gating. Identifies the active root uptake window per tree per night. Operates on deltas and directional patterns within the same sensor over time — systematic calibration error cancels out. - **Coupled-constraint fertigation**: Fertilizer recommendations respect hard chemical compatibility constraints (calcium nitrate never co-applied with phosphates, which would precipitate as calcium phosphate) before any AI output is generated. - **Per-tree longitudinal dataset**: Each node builds a continuous history tied to a specific tree. The data moat compounds over time and cannot be retroactively replicated — analogous to Stripe's fraud-signal advantage. - **Open hardware, proprietary intelligence**: Hardware Gerbers/BOM are published so customers self-manufacture via JLCPCB. All agronomic logic runs server-side. Open hardware reinforces rather than undermines the intelligence moat. - **50MP visual pest detection**: High-resolution fixed and roving cameras image insect traps, flowering panicles, developing fruit, and individual leaves. 50MP resolution is deliberate — it resolves individual insects on a sticky trap and mm-scale leaf lesions, enabling detection at low population density before damage thresholds are crossed. - **Bio-pesticide protocol layer**: When pest or pathogen pressure is detected visually, the recommendation engine applies bio-pesticide protocols first, reserves chemical intervention for confirmed escalation, and localises treatment to the affected tree or zone. - **pH oscillation management**: pH is deliberately cycled between ~5.5 and 6.3 using calcium nitrate (raises pH) and acidic nutrient mix (lowers). The AI layer tracks cycle phase state explicitly. - **Neighbourhood baseline principle**: Trees in the same block share soil type, microclimate, and fertigation history. A newly instrumented tree is interpreted against the baseline of reference trees, compressing calibration from weeks to days. Hardware does not scale linearly with tree count. --- ## Architecture summary - **Sensor nodes**: ESP32-S3-WROOM-1U, RS-485 Modbus at 4800 or 9600 baud, multiple soil sensor families (ZTS/SN, JXCT, DFRobot). Solar-powered (20W panel, 12V, LiFePO4 6Ah). External U.FL antenna for above-canopy WiFi. - **Field network**: Hub-and-spoke WiFi mesh (TP-Link repeaters). Raspberry Pi gateway running Python controller, polling Cloudflare KV every 2 seconds. - **Fertigation control**: 7 Tuya devices controlled via tinytuya LAN protocol — bore pump, dosing pumps, valve timers. - **Backend**: Cloudflare Workers / D1 / KV / R2 / Pages. Edge-deployed globally. - **AI layer**: Dual-model. Claude Haiku for continuous sensor auditing; Claude Sonnet for deep agronomic analysis and per-tree recommendations. --- ## Competitive positioning - **vs. Fasal (Wolkus Technology Solutions)**: Fasal deploys field-level soil moisture and temperature sensors. Their EC/pH sensor is an inline fertigation bench sensor measuring outgoing water — not in-soil. No per-tree longitudinal intelligence, no nocturnal uptake window detection, no visual pest intelligence. - **vs. Netafim NMC**: Irrigation scheduling automation. No per-tree soil intelligence, no uptake signal detection. - **vs. generic farm apps**: Field-level averages, calendar-based scheduling. No longitudinal per-tree memory, no physiological signal detection, no visual pest intelligence. --- ## Market Current deployment: Alphonso mango, Gujarat, India. Architecture is crop-agnostic. Identified expansion markets: citrus, pomegranate, wine grapes (largest global market opportunity, highest willingness to pay). --- ## What is not indexed - /admin/ — internal dashboard, authentication required - /live/ — real-time sensor dashboard, JS-rendered, session-authenticated - /api/ — backend API endpoints - /login/ — authentication page --- ## Contact admin@sankhyafarms.com Sankhya Ventures LLC (Delaware)