reogento/puchinaai

How reogento/puchinaai works

PuchinaAI is a research and operational system for forecasting road frost-heave and deformation risk at 30-, 60-, and 90-day horizons. Its research workflow loads or generates telemetry, derives physical freeze–thaw features, builds chronological horizon-specific datasets, trains models, and evaluates risk and alarm thresholds. Engineers can also run a separate CLI inference workflow using saved models, with batch predictions and operational recommendations. The graph follows documented and indexed relationships; some internal links are omitted where excerpts do not establish the caller or data flow.

Main entry points: main.py Research CLI · predict.py Inference CLI · data_loader.py Dataset loader · dataset_builder.py Horizon datasets

Components 16

Telemetry and labeling

Synthetic telemetry src/data_generator.py
Dataset loader src/data_loader.py
Defect labeling src/label_dataset.py

Research pipeline

Research CLI main.py
Physical features src/features.py
Horizon datasets src/dataset_builder.py

Risk models

RF and XGBoost src/models/baselines.py
Hybrid neural model src/models/hybrid_nn.py

Operational inference

Inference CLI predict.py
Saved models saved_models
Prediction and research outputs results

Evaluation and interpretation

Model evaluation src/evaluate.py
SHAP interpretation src/interpretation.py
Alarm thresholds src/threshold_analysis.py
Physical plausibility checks run_plausibility_checks.py

Other components

Road engineer

Connections 17

Road engineer → Research CLI: runs research README.md
Road engineer → Inference CLI: requests forecast README.md
Research CLI → Dataset loader: loads data main.py
Dataset loader → Defect labeling: labels telemetry src/data_loader.py
Dataset loader → Synthetic telemetry: generates fallback src/data_loader.py
Research CLI → Horizon datasets: builds datasets main.py
Horizon datasets → Physical features: computes features src/dataset_builder.py
Research CLI → RF and XGBoost: trains baselines main.py
Research CLI → Hybrid neural model: trains hybrid main.py
Research CLI → Model evaluation: evaluates predictions main.py
Research CLI → SHAP interpretation: interprets models main.py
Research CLI → Alarm thresholds: analyzes alarms main.py
Research CLI → Physical plausibility checks: checks plausibility
Inference CLI → Physical features: computes features predict.py
Inference CLI → Saved models: loads model predict.py
Inference CLI → Prediction and research outputs: exports forecasts README.md
Research CLI → Prediction and research outputs: writes metrics main.py

Drawn by GitDiagram on November 11, 2026 from the repository's file tree, README and source files. Read as Markdown ·