ZakiANK04/AgroVisor---Algeria-2.0

# 🌾 AgroVisor Algeria ### AI crop advisory for Algerian farmers — in Arabic, French, and English *Three XGBoost models turn a farm's region, soil, and season into a price forecast, a yield estimate, and an oversupply risk score — then rank which crops are actually worth planting.* -success?style=flat-square) --- ## The problem An Algerian smallholder decides what to plant with three unknowns: what the crop will sell for at harvest, how much land will actually yield, and whether every other farm in the wilaya is planting the same thing. Get the third one wrong and a good harvest becomes a price collapse. AgroVisor makes all three predictable. A farmer registers a farm — location, size, soil type — and gets a ranked list of crops with a forecast price in DA/kg, an expected yield in tons/ha, and an oversupply risk percentage, plus contextual advice generated in their own language. --- ## Why this project is worth a look | | | |---|---| | **Three models, one decision** | Risk classification, price regression, and yield regression are trained separately and combined into a single recommendation score with a risk threshold — not one model asked to do everything | | **Genuine RTL support** | Full Arabic localisation with right-to-left layout, including translated region names, crop names, and soil types — not just a string file swap | | **Region-aware defaults** | Soil type auto-detects from wilaya via `region_soil_mapping.csv`, so a farmer who doesn't know their soil classification can still get a recommendation | | **The feedback loop is built in** | Predictions are written back to `model_results.csv` with duplicate prevention, creating a retraining corpus from real usage | | **LLM as a layer, not the product** | OpenAI generates the human-readable advice *around* the model output. The numbers come from XGBoost, not from a language model | --- ## 🏗️ Architecture ```mermaid flowchart LR subgraph FE["React 19 + Vite"] L["Landing"] A["Auth — J …