My background is in mechanical engineering, vibration and modal analysis, and equipment diagnostics. I am independently studying numerical methods, statistics, and machine learning through personal projects and documented case studies.
My public work focuses on structural dynamics, finite element methods, signal processing, and data-driven experimentation. The repositories connect mathematical models with numerical implementations and examine their agreement with analytical solutions, synthetic ground truth, or public experimental datasets.
| Project | Scope and validation evidence | Explore |
|---|---|---|
| mech-design-study | Beam stress, deflection, vibration, and steady-state heat calculations; beam FEM checks against analytical solutions; plane-stress element comparisons and mesh-convergence studies using manufactured solutions. Educational models with explicit unevaluated requirements. | FEM verification · Mesh convergence |
| modal-analysis-portfolio | Experimental modal-analysis methods: LSCF/CMIF pole detection, shared-pole MIMO fitting, and mode-shape visualization. Published benchmarks use synthetic frequency-response data with known modal parameters. | Streamlit demo · Validation scripts |
| bearing-diagnostics | FFT/PSD and envelope-analysis checks against NumPy/SciPy, known-fault tests on CWRU data, and degradation detection on NASA IMS run-to-failure data. Independent-run transfer results include both successes and failures. | Case study |
| bayesopt-process | Gaussian-process regression and acquisition functions implemented in NumPy and checked against reference implementations. Sequential experiment selection, noise sensitivity, and calibration are studied in a fixed-pool virtual experiment using public concrete-strength data. | Case study |
| geo-change-detection | Sentinel-2 wildfire-change detection compared with Copernicus EMS reference maps. Core numerical operations are checked against established libraries; evaluation examines how reference definitions affect reported metrics. | Case study |
| agent-viz | MLflow-backed experiment records, case-level comparisons, and linked decision evidence. A shared visualization and feedback interface supports review by humans and AI agents; a synthetic demo is included. | Repository and setup |
The modal-analysis demo may need time to wake after inactivity. Case-study reports can be read without downloading the source datasets; recomputation requires the inputs and dependencies documented in each repository.
Across these projects, I emphasize:
- Numerical verification: compare implementations with analytical solutions, known-answer cases, or established libraries, with assumptions and tolerances stated.
- Controlled comparisons: define hypotheses, baselines, validation splits, and decision criteria before interpreting improvements.
- Evaluation context: report metrics alongside their reference definitions, evaluation populations, uncertainty, and applicable conditions.
- Traceable results: retain experiment settings, comparison records, failed hypotheses, and corrections.
- Limits of evidence: distinguish implementation checks, synthetic benchmarks, public-data validation, and transfer to new conditions.
These repositories are personal development and learning projects, with AI assistance used for implementation and documentation. Numerical agreement and benchmark performance are evidence within the stated test conditions; application to physical equipment or new datasets requires further validation.
- House Prices — outlier handling and nested cross-validation for model selection, with fold-level comparison records.
- Store Sales — 16-day forecasting, time-series cross-validation, and comparisons between local validation and recorded public leaderboard results.
- Titanic — pre-registered comparisons, subgroup error analysis, uncertainty, and documented disagreements between cross-validation and public scores.
Visualization, learning tools, and personal applications
- Data exploration: similarity-radar.
- Language learning: enja-reader, scenelex, and ai-english-conversation-tutor.
- Personal and interactive tools: kajiflow, math-painting, and forza-horizon-telemetry-monitor.
- Trade-history analysis and screening: trade-statistics and jp-stock-screener.