Selected systems that show how I build when the work has to ship.
A three-tier nightly ETL that folds four operational systems into one 78-column fact table.
| Source | System | Role |
|---|---|---|
| Internal API | REST API | events |
| Maximo | ERP | assets |
| Billing orders | Billing | usage |
| LDAP | Directory | identity |
150K records a day since November 2025, across DEV / QA / PROD.
A studio that cleans hidden or malformed Unicode, inside one shared catalog of PDF and text tools.
| App | Role | Deploy |
|---|---|---|
| studio | hub / catalog | public |
| proofmark-pdf | PDF suite | private |
| text-cleaner | text sanitizer | private |
Fun days, real public holidays, and on-this-day history for Canada, the US, and Mexico, in one fast place.
a coverage gate fails the build if holiday data collapses again.
A fix to a production pipeline that was misreporting order value in both directions: it restored $4.1M that was under-counted, including $2.7M across 8,355 missing orders, and removed $22.7M that was double-counted.
The approved high-level design for a four-layer BigQuery warehouse: a landing layer, core data models with declared grains and schema-contract tests, data marts, and a semantic layer on top, plus the approach for unpacking nested orders. The build has not started.
A single MicroStrategy Intelligent Cube, 77 attributes and 12 metrics, built on data from four separate systems, plus the KPI dashboard data products released from Dev to Prod on top of it.
A Python routine that measures how long each stage of an agent work cycle takes, across events that do not always arrive in order.
The investigation and repair work that found why historical data was missing and wrong, then recovered 28K+ silently dropped records and stopped it from happening again.
A rework of how a reporting job reads a 23-million-row device inventory table on Teradata. The old query joined everything and filtered later; the new one filters the lookups first, pushes predicates down into them, and accounts for key skew.
A transformation step that reconciles the same fact (a customer name, a product, a ticket number) when four systems each spell it differently, and picks the single best value.
The pipelines and cloud setup that land MongoDB Atlas collections in BigQuery and check that the copy matches. Built as code: Terraform, Cloud Composer DAGs, and Cloud Run ingestion jobs deployed through GitLab CI/CD, with watermark-based change data capture and MERGE upserts.
An open-source web platform researchers use to explore, simulate, and analyze biological data without writing their own tooling from scratch: Shiny and React dashboards over Python and R services, Docker microservices, and REST and GraphQL APIs.
A multi-omics ETL on HPC clusters that pulls cancer data together from several public sources, quality-checks it (automated K-Means and DBSCAN anomaly checks in CI/CD), and runs machine-learning models (SVM-RFE, random forest) to surface candidate biomarkers.
A tool that turns a Markdown résumé, with a live preview, into clean DOCX, PDF, TXT, and a JSON QA report, and can also check an existing PDF or DOCX résumé for how well applicant-tracking parsers will read it.
Live app ↗A system that audits all my repositories once a week, reading commit history and repo state, and writes a dated report, then keeps my tools and config in sync across my Windows and Mac machines automatically.
A pipeline that loads NYC taxi-trip Parquet files into Postgres, then transforms them in medallion layers (bronze → silver → gold) into star-schema fact and dimension marts with dbt. dbt tests check every build, Airflow on Docker Compose orchestrates it, and reloads are idempotent.
Repository ↗A local-first web app for planning SAS data-integration job flows: which jobs run, in which environment lane, in what order. Dependencies are cycle-checked, and a plan exports as a runbook, CSV, SVG, or JSON.
Repository ↗Live app ↗An R package that takes 16S and metagenomic data from raw tables to answers: DESeq2 or edgeR differential abundance, random-forest classification, and an interactive Shiny app, so researchers stop writing the same analysis code by hand.
Repository ↗Live app ↗A video pipeline: upload a clip, a Node/Express API hands it to a Python YOLOv5/OpenCV detector, a configurable rule flags anomalies (such as too many people in frame), and each alert lands in PostgreSQL on AWS RDS, all running on EC2.
Repository ↗A free learning platform for cell and molecular biology: interactive tutorials, quizzes, and real pathway simulations you can tweak and watch update live. A free alternative to Cell Collective.
Docs ↗A job board that gathers technology roles at nonprofits into one searchable place, with automated Greenhouse adapters, an operator-gated refresh, and honesty about which sources are still manual.
Live app ↗The commercial twin of Signal for Good: the same job-board engine, but its source adapters cover ten different corporate hiring systems, and repeat postings collapse into one.
Live app ↗A native iOS app (React Native / Expo) that suggests duas and adhkar from context (prayer times, time of day, weather) and shows the source of every one, with on-device prayer times and a Qibla compass. Local-first: no account and no tracking.
Live app ↗App Store ↗This website. A field-manual-styled portfolio that ties the story, the proof, and a gated deeper archive together instead of scattering them across disconnected pages.
Live app ↗Smaller builds, experiments, and works in progress.