I am a machine learning engineer, computational biologist, and systems designer working where clinical informatics, cybernetics, and information theory meet. My day job is building high-stakes data and ML systems for oncology and real-world evidence; my longer-horizon work is about treating those systems as goal-directed, feedback-rich processes rather than mere data plumbing.

Formally, my background combines bioinformatics & computational biology, computer science, and data infrastructure engineering. Conceptually, I draw a line from early cybernetics (control and communication in organisms and machines) through information theory (information as the resolution of uncertainty) to modern AI and multi-scale cognition. My focus is making that lineage concrete in code: ontologies become vectors; feedback loops become APIs; evaluation becomes a first-class artifact.

• Cybernetics & information flow

  • Model clinical platforms as feedback systems: sensors (EHR, labs, genomics), controllers (mapping engines, policies), actuators (dashboards, decision support).
  • Treat pipelines as communication channels with noise, capacity, and distortion; design for graceful degradation instead of silent failure.

• Vectorized ontologies & representation geometry

  • Embed NCIt, SNOMED CT, UMLS, RxNorm, FHIR value sets, and OBO ontologies into vector spaces; analyze manifold structure, capacity, and community detection (Leiden/Louvain).
  • Build mapping engines that combine approximate nearest neighbors, lexical features, and domain constraints to align FHIR resources and procedural codes to ontology terms.

• Clinical data systems & governance

  • Architect federated data meshes: per-site marts and warehouses backed by relational + vector stores, coordinated by a mesh hub with policy, lineage, and evaluation.
  • Emphasize auditability and epistemic humility: every mapping, score, and model decision should be traceable, inspectable, and falsifiable.

• Working style

  • Kanban-driven development, small reviewable PRs, and CI that enforces formatting, linting, tests, and documentation.
  • Preference for strong types and explicit invariants (Rust / typed schemas) in pipelines that must be correct for years, not just demos.

• Data-First Procedural Semantics (DFPS) / clinical platform work

  • Multi-crate Rust workspace for:
    • FHIR bundle ingestion, validation, and normalization.
    • Ontology-aware mapping of service requests, procedures, and observations to NCIt and related vocabularies via vector backends (FAISS / pgvector / similar).
    • Mesh-style orchestration across local datamarts, analytics marts, and shared governance layers.

• Mapping evaluation & information-theoretic probing

  • CLIs to build, query, and introspect vector indexes; tools to compare lexical vs. embedding-based vs. hybrid matching strategies.
  • Analyze error modes as information-processing failures: ambiguous codes, underspecified contexts, brittle embeddings, and graph pathologies.

• Frontends & inspection tools

  • Next.js + Tailwind + ShadCN UIs for:
    • Inspecting mapping neighborhoods (top-k candidates, confidence scores, lexical/semantic evidence).
    • Visualizing ontology graphs, local manifolds, and evaluation metrics in ways that clinicians and data stewards can actually reason about.

• Languages & ecosystems

  • Rust for domain models, ingestion pipelines, mapping engines, governance / mesh services.
  • Python for experimentation, data analysis, evaluation harnesses, and research prototypes.
  • TypeScript / React / Next.js for visual analytics, operator consoles, and developer tooling.

• Data & infra

  • PostgreSQL + SQLx for relational cores; dimensional datamarts / warehouses for downstream analytics.
  • Vector stores (FAISS, pgvector-style backends, ANN indices) as an explicit ontology layer, not an afterthought.
  • Containers and IaC (Docker, Terraform-style tooling) with GitHub Actions CI that runs fmt, lint, test, and doc checks.

• Design principles

  • Domain-driven structure: domain (semantics), platform (infrastructure), app (interfaces), each with narrow, testable contracts.
  • Extensive CLI entry points (e.g., build_vector_index, map_bundles, map_codes, eval_mapping, load_datamart, validate_fhir) so experiments and pipelines are scripted, versioned, and repeatable.
  • Treat metrics, logs, and traces as feedback signals for a cybernetic system rather than mere observability garnish.

• Programming

  • Primary: Rust, Python, TypeScript
  • Also: SQL, Bash, occasional JVM/web tooling when interfacing with legacy systems
  • Tooling: cargo, poetry/pip, Node/Bun, modern linters/formatters, GitHub Actions / similar CI.

• Communication

  • Write design docs, evaluation reports, and governance notes that tie together code, data, and outcomes.
  • Strong bias toward:
    • Declaring assumptions and failure modes up front.
    • Making epistemic status explicit (“measured”, “estimated”, “hypothesized”).
    • Translating between engineers, clinicians, data stewards, and leadership.

“Information is information, not matter or energy.”
— Norbert Wiener

“Information is the resolution of uncertainty.”
— Claude Shannon

“Artificial intelligence is the science and engineering of making intelligent machines, especially intelligent computer programs.”
— John McCarthy

“Novel beings, novel goals.”
— Michael Levin