Victoria G. Walter

An R Shiny analytics platform backed by a SQL Server database (C2DB), supporting cost benchmarking and performance evaluation of multi-billion dollar NNSA capital acquisition programs. It serves 100+ federal cost estimators and program analysts. I have worked on it for five years and have led the project since October.

The platform exists to do three things: consolidate dirty, disparate source data into one accurate and documented store with business logic encoded in an automated validation pipeline; replace siloed Excel models with interactive, version-controlled tools inside the dashboard; and let analysts build and export their own queries rather than waiting on someone else to pull the data. All three serve the same end — trusted, tailored data that lets cost estimators reach defensible conclusions faster.

Role

Project lead since Oct.; senior contributor prior

Team

3–5, several part-time

Stack

R Shiny · SQL Server · bslib · DBI/dbplyr · pointblank · Azure

• Led the v4.0 production release: set release strategy through repeated engagement with stakeholders across teams, ran the delivery sprint, owned build of a new analytical model, and reviewed and approved the team's work into production
• Designed the SQL Server governance layer — triggers, change log, and SQL views — including a data dictionary maintained as an extended property, so documentation lives with the data rather than beside it
• Established version control practice for both CAAT and IIP: set up and connected the GitHub repositories, wrote the documentation covering account setup, cloning, and contribution workflow, and administer access permissions across both projects
• Built several of the dashboard's analytical visualizations and one of its models, designed against UI/UX best practice — consistent interaction patterns, deliberate information hierarchy, and layouts organized around how analysts work rather than how the data is stored
• Leading the ongoing migration of the database and Posit Connect server off on-premise hardware and onto Azure
• Wrote the in-depth training guide, ran dashboard trainings, and led custom demos built around each incoming stakeholder group's workflows
• Led onboarding and knowledge sharing across the team, including the technical documentation new contributors need to become productive

• Contributed to the database redesign into a snowflake schema
• Helped define business logic and built a portion of the automated validation suite using pointblank
• Wrote data cleaning and loading functions, and contributed to the custom R package behind the ETL pipeline
• Migrating the legacy Excel models onto the platform brought them under version control, added documentation, and let inputs flow in automatically rather than being keyed in by hand
• Adoption grew from passive recipients to active daily users, with day-over-day dashboard hits increasing more than 50%

A decision-support platform where analysts construct and optimize portfolios of capital projects, adjusting start years, durations, and funding levels. The problem it addresses is a three-way constraint: a portfolio has to be affordable against the budget available, executable given real scheduling and capacity limits, and still sufficient to meet mission requirements.

My contribution is the data layer, IIPDB. Most of its complexity exists to give users flexibility. Analysts can select from a wide range of cleaned, documented inputs when configuring a simulation, so the database holds those inputs for all 225+ capital projects in the portfolio — three to five configurable inputs per project, each carrying one to three data points — and persists every scenario a user builds.

Role

Data layer owner

Stack

SQLite → PostgreSQL · R package dev · medallion architecture

• Architected and maintain IIPDB — a database and custom R package pipeline following the medallion model: raw ingestion → silver standardization → gold transformation
• Designed the schema — tables, relationships, constraints — around the access patterns the optimization models actually use, rather than around the shape of the source data
• Silver layer standardizes disparate sources into consistent, query-ready tables, separating ingestion concerns from analytical logic
• Gold layer applies custom R modeling functions that encode domain business logic and pre-compute results, keeping computation out of the Shiny frontend
• Currently migrating the store from SQLite to PostgreSQL as concurrent scenario writes outgrow what SQLite handles comfortably

A county-level analysis of Pennsylvania's economic and political realignment across five election cycles. The prevailing account attributes the shift to manufacturing job losses concentrated around 2016. The data does not support that timeline: manufacturing had been leaving the state for decades, and the erosion of working and middle class economic position was correspondingly gradual. What changed more recently was not employment but pressure — households already weakened by long-run wage stagnation meeting steeply rising costs of living.

Role

Sole author

Stack

R Shiny · tidycensus · leaflet · Census/ACS

Status

Live dashboard ↗

• Built an economic indicator panel across all 67 counties from Census and ACS sources, tracking wage growth, cost burden, and manufacturing employment over a multi-decade window rather than a single cycle
• Classified counties by five-election flip pattern and tested the long-horizon stagnation thesis against outcomes
• Separated long-run structural decline from recent cost pressure, so the two effects could be examined independently rather than collapsed into one narrative
• Constructed a forward-looking targeting score identifying highest-return counties for 2026 and 2028
• Deployed as an interactive dashboard where users adjust indicator weights and watch the ranking respond

The campaign's matching-funds claims depended on donation data reaching the NYC Campaign Finance Board both accurately and on schedule. I built the pipeline that delivered it — running unattended against a fixed statutory deadline, in a context where a reconciliation error carries immediate financial consequence rather than a correction in next quarter's report.

Role

Bluebonnet Data fellow

Stack

Python · Docker · ActBlue

• Built automated Python and Docker pipelines reporting ActBlue donation data to the NYC Campaign Finance Board
• Eliminated 20+ hours of manual reconciliation per week, and removed the error class manual entry introduces
• Supported $6M+ in matched donations through reliable automated reporting

An outside political consulting firm circulated a report indicating that support among one demographic group was declining week over week. I analyzed census and voter file data to examine the same question over a longer horizon, and what the underlying trends showed pointed in the other direction.

Separately, field organizers were spending hours each week hand-entering records of organizing and fundraising activity. I automated that reporting.

Role

Bluebonnet Data fellow

Stack

R · Census/ACS · VAN/VoteBuilder

• Analyzed census and voter file data in R to examine demographic support trends over time, testing a widely circulated claim against the underlying evidence
• Used those trends to inform precinct-level outreach strategy
• Automated internal reporting on organizing and fundraising activity, eliminating hours of weekly manual data entry for field organizers