Machine learning model training authority — classifier and regressor training with scikit-learn and PyTorch, feature engineering, cross-validation, hyperparameter tuning, fine-tuning transformer models with HuggingFace, dataset splits, loss functions, learning rate schedules, and reproducible training runs.
MLOps infrastructure engineering covering model serving (vLLM, TorchServe, Triton), feature stores, model registries (MLflow, W&B), A/B and shadow deployments, drift detection, Prometheus/Grafana ML metrics, Kubeflow Pipelines, and Airflow DAGs for end-to-end training and deployment pipelines.
Running multiple AI coding agents (Claude Code, Codex, OpenClaw) in isolated containers that share a single repo clone via /opt bind mounts, communicate over SSH, and avoid redundant git clones — container topology, volume strategy, inter-agent SSH, and coordination patterns.
Partner skill compiler authority — synthesize grounded public evidence into a reusable partner prompt, persona profile, and SKILL.md with worldview, operating style, audience, offers, voice anchors, and strict anti-hallucination constraints.
Public persona research authority — discover, rank, and curate public sources for a person across websites, YouTube, podcasts, newsletters, LinkedIn, and social profiles while avoiding name collisions, wrong-identity contamination, and low-signal pages.
SI units, CODATA constants, dimensional analysis, and rigorous unit conversion. Use when extracting, normalizing, or computing physical quantities (energy, frequency, length, mass, time, temperature). Handles eV/keV/MeV/GeV/TeV, Hz/kHz/MHz/GHz/THz, meter/cm/Å/fm, kg/g/u, second/ms/μs/ns/fs, Kelvin/eV-temperature…
Routes AI tasks to dynamic LLM-generated skills via an orbital classifier. Runs locally from the meridian-skills-mcp npm package. Needs 3 environment variables to run.