Skip to content
View jacobByron-McKay's full-sized avatar

Block or report jacobByron-McKay

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
jacobByron-McKay/README.md

Jacob Byron-McKay

Senior Data Scientist with 6+ years across the New Zealand public sector (health and analytics) - working end-to-end from messy operational data through to production analytics, machine learning, and decision-ready reporting. Māori (Ngāti Apa). Strong in R, SQL and Power BI, and in Python end-to-end across the projects below, with hands-on work in applied ML/NLP, retrieval-augmented generation, and locally-run large language models in regulated, privacy-sensitive settings.

About these repositories

They are public-data portfolio projects that demonstrate techniques I use in professional work. Work projects use confidential data and can't be published, so these examples are built on open datasets - the methods, the rigour and the decisions are the same.

Selected work

  • clinical-guideline-rag: retrieval-augmented question answering over UK NICE clinical guidelines, running entirely locally (Ollama + Chroma): grounded answers that cite the exact recommendation and refuse when the guidelines don't cover the question, with a RAG evaluation harness for retrieval quality and faithfulness.

  • hospital-readmission-risk: an end-to-end, equity-aware 30-day hospital readmission model: leakage-aware data preparation, probability calibration, SHAP explainability, a fairness audit across demographic groups, and a FastAPI service with tests and CI.

  • smoking-cessation-effect: causal inference: the effect of quitting smoking on weight gain, estimated with g-computation, inverse-probability weighting and a doubly-robust estimator, plus overlap/balance diagnostics and an E-value sensitivity analysis.

  • flu-forecasting: short-term influenza forecasting evaluated with rolling-origin backtesting against honest baselines; a gradient-boosted model beats seasonal-naive, with a regime-aware read on the COVID-era disruption.

  • clinical-note-classifier: classic ML vs a local Llama on clinical-text classification: a numbers-backed look at when a trained model beats an LLM, and when it doesn't.

  • health-analytics-dbt: analytics engineering: a tested, documented dbt + DuckDB pipeline turning raw health records into layered staging and mart models with data-quality tests.

  • health-dashboard-powerbi: an interactive Power BI dashboard on top of those dbt marts: a star-schema model, DAX measures, and coordinated multi-page report, shipped as a text-based project (.pbip) alongside the built .pbix.

Toolbox

R (tidymodels, Shiny) · SQL & database design · Power BI (DAX, RLS) · Python (pandas, scikit-learn) · machine learning · calibration & SHAP explainability · causal inference · time-series forecasting · NLP & text classification · GenAI / LLMs (local Llama, RAG, LangChain, vector databases, evaluation) · dbt / analytics engineering · equity-aware analytics

LinkedIn

Popular repositories Loading

  1. hospital-readmission-risk hospital-readmission-risk Public

    End-to-end, equity-aware 30-day hospital readmission risk model on public UCI data

    Python

  2. jacobByron-McKay jacobByron-McKay Public

    Profile

  3. clinical-note-classifier clinical-note-classifier Public

    Classic ML vs a local Llama for clinical-text classification

    Python

  4. clinical-guideline-rag clinical-guideline-rag Public

    Retrieval-augmented Q&A over UK NICE clinical guidelines - grounded, cited, refuses out-of-scope, runs fully locally via Ollama

    Python

  5. flu-forecasting flu-forecasting Public

    Short-term US influenza forecasting (CDC ILINet) with rolling-origin backtesting against honest baselines

    Python

  6. smoking-cessation-effect smoking-cessation-effect Public

    Causal inference: the effect of smoking cessation on weight gain (NHEFS) via g-computation, IPW and doubly-robust estimation, with balance and sensitivity checks

    Python