Portfolio

Projects

Data platforms, statistical analysis, reporting systems, and full-stack applications built to be used and inspected.

Personal · Data Platform
May 2026 – Present

Indigo Circuit

A competitive Pokémon TCG analytics platform that uses a DuckDB and dbt pipeline to produce live rankings, meta analysis, and matchup estimates.

  • Nightly pipeline ingests API and scraped tournament data, recomputes Glicko-2 ratings, rebuilds dbt marts, and validates a shadow DuckDB database before atomic promotion.
  • Bayesian EV model shrinks sparse card effects toward archetype baselines and reports 90% credible intervals alongside field-aware rankings.
PythonDuckDBdbtFlask
Worth noting: The validation gate keeps the last known-good dataset live while the next rebuild proves itself. Correctness is a deployment concern, not only a modeling concern.
Open App →
Independent Analysis · Health
May 2026 – Present

Social Vulnerability & Health Outcomes in America

An analysis of relationships between CDC vulnerability indicators and health outcomes across approximately 73,000 census tracts.

  • Analyzed 16 CDC vulnerability indicators and 36 health outcomes using correlations, state rankings, scatterplots, and a national choropleth.
  • Kept causal claims appropriately bounded with literature-grounded caveats rather than treating correlations as explanations.
PythonStatistical analysispandasgeopandas
Worth noting: Public datasets become useful only after their geographic grain is made comparable. The methodological caveats are part of the result.
View Analysis →
Personal · Full-Stack Web App
May 2026 – Present

RareDex

A card-scanning application that records scans, recognition failures, price snapshots, and user collections in PostgreSQL.

  • Built a layered identification pipeline using ranked image preprocessing, collector-number OCR, and a 19,601-entry perceptual-hash fallback.
  • Used SQLAlchemy and Alembic to manage persistence and schema changes across the application.
FlaskPostgreSQLSQLAlchemyOpenCV
Worth noting: The pipeline escalates through bounded recognition methods before giving up, making recognition failures observable rather than silently confident.
Open App →
Personal · Robotics Simulation
August 2026 – Present

C-1N

An evolving six-legged MuJoCo robot whose named checkpoints preserve the progression from coordinated failure to understood mechanics and better locomotion.

  • C-1N // 01 · SHUFFLE preserves the current phase-shared tripod gait, torso-orientation feedback, foot-contact checks, and failure to sustain forward walking.
  • The live browser MuJoCo/WebAssembly artifact keeps historical checkpoints inspectable so later control changes can be compared against the same robot lineage.
PythonMuJoCoWebAssemblyControl
Worth noting: C-1N is deliberately longitudinal. A checkpoint advances only when an experiment produces a capability or understood failure worth preserving; presentation polish alone does not increment it.
Inspect C-1N →
Data Systems · Microsoft
Jul 2021 – Jul 2024

SSAS Multidimensional → Tabular Migration

Migrated an internal financial reporting platform to Tabular with zero downtime, owning implementation, telemetry, validation, and operational support.

  • Extended automated financial-value reconciliation between the legacy and replacement systems to preserve correctness during the migration.
  • Reduced representative filtering latency by up to 200x and user-visible Excel loading time by approximately 4x.
SSAS TabularSQLAzureTelemetry
Worth noting: A migration earns trust when the new system can prove parity with the old one while users continue working.
Stakeholder Analytics · UVA
May 2025 – Mar 2026

Decision-Support Dashboards

Owned Tableau analytics products for institutional stakeholders from discovery through implementation, deployment, and ongoing support.

  • Translated ambiguous operational questions into SQL-backed metric definitions and decision-support dashboards.
  • Contributed to telemetry-informed optimization work that improved organization-wide dashboard load times by approximately 20%.
TableauSQLTelemetryStakeholder delivery
Worth noting: The useful work begins before the visualization: first make the question and metric definition precise enough to support a decision.