Projects-Portfolio

Richard Seyeau

Data analytics, business intelligence, and IT service management professional. Open to remote and hybrid roles in data, AI, IT, and tech — including the Canadian federal government and data centre operations.

CI

About

I work across the data lifecycle: cleaning and validating data, analysing it with statistics and machine-learning methods, and presenting it through dashboards and reports that non-technical stakeholders can act on. I hold professional certificates in data analytics (Google, IBM), statistics (Wolfram), Power BI (Microsoft), and ITIL 4 service management, and I keep building on that foundation through applied projects — the ones in this repo are documented, tested, and reproducible.

Skills

Area Skills
Data analysis Exploratory analysis, data cleaning, transformation, validation, modelling, predictive analytics, regression
BI & reporting Power BI, Tableau, Excel, IBM Cognos Analytics, dashboards, report writing, data storytelling
Programming & databases Python, Pandas, NumPy, Polars, DuckDB, SQL, R, PostgreSQL, MySQL, MongoDB, GraphQL, Jupyter
Statistics & ML Descriptive and inferential statistics, probability, hypothesis testing, ANOVA, regression, supervised and unsupervised learning, model interpretation
IT service management ITIL 4 Foundation — service value system, governance, continual improvement
Data quality & governance Metadata management, documentation, integrity checks, privacy-aware handling of information
Tools Visual Studio Code, Git/GitHub, RStudio, Google Colab, Power Query, Microsoft Access

Certifications

Experience highlights

Preview Project
Exoplanet classifier report Exoplanet transit classifier — 0.946 cross-validated ROC-AUC on 5,000+ Kepler light curves, zero false positives on the held-out test set. The same transit-hunting pattern I do on Zooniverse, made reproducible.
Fine-tuning loss curve LLM fine-tuning workflow — LoRA fine-tuning on a laptop CPU, validation perplexity 41.4 → 22.8. Config-driven, tested, documented end to end.
Infrastructure health monitor report Infrastructure health monitor — SLO tracking with four check types, transition and breach alerts, self-contained HTML reports. Live demo caught a flaky service and a crashed worker.
Disk failure prediction report Disk failure prediction — flags drives at risk of imminent failure from S.M.A.R.T. telemetry: gradient boosting on a 60,000-drive sample of Backblaze’s real Q1 2024 data, 0.9999 held-out ROC-AUC, 0.994 precision / 0.970 recall, with recall above 0.98 up to 30 days before failure and an honest false-alarm analysis.
Grocery inflation dashboard Canadian grocery inflation — ten years of StatsCan CPI data showing groceries outrunning overall inflation (+43.4% vs +36.0%), with an EDA notebook and self-contained dashboard.
Unemployment chart from the warehouse queries Canadian labour market analysis — Statistics Canada labour force and population tables normalized into a SQLite warehouse, answered through a 14-query SQL catalog. New Brunswick’s unemployment gap versus Canada narrowed from 4.2 pp (1970s) to 1.1 pp (2020s).
Labour market explorer dashboard Canadian labour market explorer — the warehouse’s interactive companion: R computes 50 years of labour-market statistics and exports JSON; a zero-dependency JavaScript dashboard renders them with region and metric switching, recession shading, and hover readouts. Live on GitHub Pages.
Benchmark query time and memory chart Query engine benchmark — pandas, Polars, and DuckDB on 5M orders with cross-engine result verification. DuckDB was 5–28× faster than pandas on most queries with ~2.2× less peak memory; Polars won load time and group-bys.
Solar system positions viewer Solar system positions viewer — an interactive 2D scatter-plot model of the solar system: every planet’s position for every month 1900–2100, computed with Astropy’s built-in ephemeris and viewable as a self-contained HTML timeline on GitHub Pages.
Seedworlds generated map Seedworlds — one word grows an entire world: a constructed language, a map with biomes and rivers, six creatures, and a creation myth — all consistent with each other and deterministic from the seed, so the same word always grows the same world. Pure client-side JavaScript, live on GitHub Pages.
Paradox Engine interface Paradox Engine — four browser instruments that break your own reasoning: an interrogator that builds justification trees and flags circularity, a fallacy foundry that highlights twelve classic fallacies inline, a Kripke-style self-reference lab that evaluates networks of self-referential sentences (the liar → PARADOX), and a preference lab that catches Condorcet cycles in your own pairwise choices. Pure client-side JavaScript, live on GitHub Pages.

Repo layout

Folder Contents
data/ Data pipelines, analysis, and reporting — Canadian grocery inflation project, Canadian labour market analysis, Canadian labour market explorer
ai-ml/ Machine learning and AI applications — LLM fine-tuning workflow, exoplanet transit classifier
it-ops/ Infrastructure, automation, and operations — infrastructure health monitor, disk failure prediction
experiments/ Prototypes, tools, and personal projects — query engine benchmark, solar system positions viewer, seedworlds, paradox engine
docs/ Project standards and skills matrix

Contact

This site is open source. Improve this page.