AgriNexa

## 🧪 Core Algorithms & Decision Engines ### 1. Agentic Orchestration & Intent Classification AgriNexa employs **CrewAI** to model multi-agent cognitive flows, falling back to a structured native LLM router if CrewAI execution fails. * **Tool & Intent Router (`ToolRouterService`)**: Classifies query inputs into active tool pipelines (`weather`, `agent`, `rag`, `llm`) and maps tasks to six key agricultural intents: `weather`, `market`, `soil`, `pest`, `irrigation`, and `crop`. * **Multi-Agent Collaboration**: * **Farm Operations Planner**: Translates location, crop, growth stage, weather, and market conditions into structured daily tasks. * **Agri Risk Analyst**: Cross-checks operations against chemical safety boundaries and extreme weather forecasts. * **Farmer Advisory Writer**: Condenses findings into plain-language, high-impact advisories (maximum 140 words). * **Confidence Scoring Engine**: Computes reliability indexes (`low`, `medium`, `high`) based on the availability and precision of geocoding coordinates, live weather forecasts, crop growth records, and market prices. $$\text{Confidence Score} = (S_{\text{weather}} \times 0.4) + (S_{\text{market}} \times 0.3) + (S_{\text{risk}} \times 0.3)$$ --- ### 2. Crop Activity Scheduling (AgriFlow Planner) The AgriFlow engine templates operations into stages, creating customized checklists for farmers. * **Stage-Gate Blueprints**: Automatically schedules activities using crop-specific blueprints (`rice`, `wheat`, `maize`, `cotton`) mapped to crop cycles: * **Land Preparation**: Basal fertilizer and leveling. * **Sowing**: Seed treatment and nursery checks. * **Vegetative**: Top-dressing splits and weed schedules. * **Flowering**: Water maintenance and pollinator protection. * **Harvest**: Drying and grain moisture checks. * **Dynamic Real-Time Alignment**: If a farmer logs a field observation indicating early flowering, the service adjusts the starting date of subsequent stages, marks prior tasks as completed/skipped, and recalculates the progress percentage. --- ### 3. Soil-Crop Suitability Matcher Recommends the top 3 optimal crops for a given land parcel using a weight-based scoring system: | Factor | Soil Type / Condition | Score Adjustment | Target Crop Impact | | :--- | :--- | :--- | :--- | | **Texture** | Clay / Alluvial | $+0.20$ / $+0.10$ | Rice, Wheat (favors water retention) | | **Texture** | Loam | $+0.15$ / $+0.10$ | Maize, Groundnut | | **Texture** | Sandy | $+0.20$ / $+0.15$ | Millet, Groundnut | | **Acidity** | pH < 5.8 | $+0.10$ / $+0.05$ | Millet, Groundnut | | **Alkalinity** | pH > 7.8 | $+0.10$ / $+0.05$ | Cotton, Wheat | | **Humidity** | Humidity > 75% | $+0.15$ | Rice | | **Location** | Region: Delta / Cauvery / Tamil Nadu | $+0.10$ / $+0.05$ | Rice, Cotton | --- ### 4. Weather-Driven Pest & Disease Risk Models Pest risk levels are determined programmatically by checking weather values against established microclimatic triggers: * **Fungal Disease Risk**: $$\text{Fungal Score} = (15^{\circ}\text{C} \le T \le 25^{\circ}\text{C} \to 40) + (\text{Humidity} > 80\% \to 30) + (\text{Rainfall} > 10\text{mm} \to 30)$$ * **Bacterial Disease Risk**: $$\text{Bacterial Score} = (\text{Humidity} > 75\% \to 50) + (\text{Rainfall} > 5\text{mm} \to 50)$$ * **Insect Activity Risk**: $$\text{Insect Score} = (22^{\circ}\text{C} \le T \le 32^{\circ}\text{C} \to 50) + (\text{Wind} < 10\text{ km/h} \to 50)$$ * **Chemical Spray Feasibility**: * **Not Feasible**: Rain $> 2\text{mm}$ (runs risk of pesticide wash-off) or Wind $> 25\text{ km/h}$ (chemical drift). * **Optimal**: Wind between $5$ and $15\text{ km/h}$ under clear skies. --- ### 5. Mandi Market Price Extraction & Storage Decision Engine * **Agmarknet API Integration**: Fetches prices from the Data.gov.in database using the API endpoint. * **Hierarchical Search Relaxation**: 1. Search by exact `State` + `District` + `Commodity`. 2. Fallback to `State` + `Market Name` (handles cases where users confuse market and district names). 3. Fallback to `State` + `Commodity` average. 4. Fallback to commodity average nationwide. 5. Distance/Fuzzy matching via Python `SequenceMatcher` to associate user coordinate inputs with closest available mandi records. * **Financial Storage Advisory**: Calculates whether storing grains in local warehouses is financially viable by checking the volatility index ($V$) against storage costs. * **Breakeven Calculation**: $$\text{Breakeven Monthly Price Growth (\%)} = \frac{\text{Storage Cost / Unit}}{\text{Current Market Price}} \times 100$$ * Recommend **Store** if forecasted price appreciation over 3–6 months exceeds storage costs and breaks even; otherwise, recommend immediate **Sell**. ---