AI Backend for Smart Agriculture

For the first release of our smart-ag platform I need you to create the AI brain that will power three corn-focused services. Nothing on the UI, mobile, or e-commerce fronts is required—just solid models wrapped in clean, well-documented REST endpoints that my frontend team can call.

Here is the exact scope.

1. Pest & Disease ID
Train a computer-vision model (YOLOv8 or better) on PlantVillage images. Corn is the only crop for now, but feel free to enrich the training mix with compatible Kaggle sources if it boosts recall. The API receives a leaf photo, returns the detected disease name plus an evidence-based treatment guideline.

2. Chemical Compliance Assistant
Using the dataset of “Allowed / Banned Agricultural Supplies” that I will share, build an LLM-powered RAG pipeline. The service must answer farmers’ free-text questions while refusing or correcting anything that contradicts the rules. Guardrails against hallucination are mandatory.

3. Yield Prediction
Create a model—XGBoost is a strong candidate—that ingests live IoT sensor readings (JSON payloads: weather, soil moisture, temperature, etc.) together with our historical records to forecast expected yield and harvest timeline for the current corn field.

Acceptance criteria
• Each model exposed through its own versioned REST endpoint with Swagger / OpenAPI docs.
• Inference latency <500 ms on a modest GPU (T4 class) or CPU fallback.
• Unit and integration tests proving API outputs on sample queries.
• Docker-compose file that spins up all three services and any required databases or vector stores.
• Read-me covering setup, fine-tuning steps, and model retraining commands.

If you are confident you can deliver production-ready code that meets the above, let’s get started.