Build AI-first AutoML platform: CSV → best model + insights

[Copilot]

Replaces the manual model-selection Flask app with a fully automated ML pipeline. Users upload a CSV and receive the best trained model, evaluation metrics, feature importances, and a plain-English explanation — no configuration required beyond selecting the target column.

Backend — backend/ (FastAPI)

• data_processor/profiler.py — dataset profiling, auto-detects problem type (classification vs regression) from target column cardinality/dtype, generates preprocessing suggestions
• data_processor/cleaner.py — null placeholder normalisation, deduplication, high-missing/constant column drops; ColumnTransformer pipeline (median imputation + StandardScaler for numeric; mode imputation + OneHotEncoder for categorical)
• model_trainer/trainer.py — trains 5 models (LogisticRegression/Ridge, RandomForest, GradientBoosting, XGBoost, LightGBM); runs Optuna tuning (20 trials) on the winner
• evaluator/evaluator.py — accuracy/F1/precision/recall for classification; RMSE/MAE/R² for regression; sorted comparison table
• explainer/explainer.py — feature importance via feature_importances_, coef magnitude, or SHAP fallback; template-based plain-English explanation
• api/routes.py — POST /api/upload, POST /api/run-pipeline/{job_id}, GET /api/results/{job_id}, GET /api/download-model/{job_id}
• app.py — FastAPI + CORS; serves React static build from frontend/build/

Frontend — frontend/ (React + TypeScript)

• Upload zone (drag-and-drop / click)
• Pipeline config panel: target column selector, Optuna tune toggle, auto-generated preprocessing plan preview
• Results dashboard: metric badges, plain-English explanation card, model comparison table, feature importance horizontal bar chart (Recharts), trained model .pkl download

Example pipeline flow

# Upload
POST /api/upload  →  { job_id, columns, detected_problem_type, preprocessing_suggestions }

# Run (tune=False skips Optuna for speed)
POST /api/run-pipeline/{job_id}  body: { target_col: "label", tune: true }
→ {
    best_model: "GradientBoosting",
    best_model_scores: { accuracy: 0.91, f1_weighted: 0.90, ... },
    comparison_table: [ { model, accuracy, f1_weighted, ... }, ... ],
    feature_importances: [ { feature, importance }, ... ],
    explanation: "Your dataset contains 5 000 rows... The best model achieved..."
  }

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.