Keegan Holt

What I did

1. 01Built an authenticated Next.js application that reads daily feeds out of SharePoint through the Microsoft Graph API, caching the data rather than the render so freshness stamps stay honest.
2. 02Modeled the season as eleven distinct questions: pacing to budget, renewals, revenue quality, per-game performance, rep activity, league context. Each one got its own module instead of one undifferentiated report.
3. 03Wrote a demand model that scores every game on the schedule and predicts attendance, with a what-if tool for testing opponent, weekday, and date-window combinations.
4. 04Ran a twelve-pass design audit that collapsed thirty ad-hoc font sizes into a nine-step scale, unified three competing color systems, and reduced nine breakpoints to three.

Outcome

One source of truth, refreshed on its own, that leadership opens instead of asking for. The morning reconciliation is gone.

What I did

1. 01Profiled every custom field across all 1.6M rows rather than a sample. Six fields that the classification plan depended on came back exactly 0% populated. The migration had created empty shells.
2. 02Traced load batches through record creation timestamps and tested the hypothesis that they mapped to source systems. Joined against the warehouse identity table and refuted it: every batch was 84–100% present in both candidate sources. Source system could not partition the file; only purchase recency could.
3. 03Identified three separate ticketing-ID namespaces being treated as one, and established that a 0% match across them was the correct expected result rather than a broken join.
4. 04Retracted a draft update file built on the wrong namespace before it reached production, and replaced the join strategy with cross-source corroboration.

Outcome

A migration cleanup that would have written surrogate keys into a production CRM was stopped by the audit that preceded it. Deliverables ship as self-contained handoff files with verification steps built in.

What I did

1. 01Mapped the real structure of the pricing platform: where the price chart actually lives, which of the nineteen price levels the pricing sheet names, and which accessible-seating codes have to move together or strand a fan segment a season behind.
2. 02Established a single verification cell, the one price level that changes across all four game ratings, so any event could be proven updated in one glance.
3. 03Replaced the hand-typed workbook with a generator that builds all 216 offers directly from the schedule, making the transcription error class structurally impossible.
4. 04Wrote a copy-template playbook that front-loads the risky edits onto one game and leaves the remaining 210 copies purely mechanical.

Outcome

A season's offers built correctly and verifiably, with the manual step that caused the original defect designed out rather than double-checked.

What I did

1. 01Built the site in Next.js and deployed it on Vercel, structured so all seven lines of care share one layout instead of drifting into seven bespoke pages.
2. 02Wrote an eleven-question intake snapshot that routes a visitor to the right starting point in about two minutes, instead of asking them to self-diagnose from a service menu.
3. 03Built a three-column comparison against typical online telehealth and traditional primary care, describing the common pattern in each category rather than naming any one company.
4. 04Put the numbers on the page: visit length, panel size and price, and cash-pay billing with no insurance, so cost is answered before anyone books.

Outcome

A site that answers the expensive questions, what it costs and who you actually see, before the consult rather than during it.

What I did

1. 01Led with the cost instead of burying it: $5.15 per month per $100,000 of actual value, given as the midpoint of a $4.87 to $5.44 range, with the reason the range exists explained in plain language.
2. 02Built a calculator so a voter enters their own home value and gets their own figure, with the surrounding copy written to make the actual-value versus assessed-value distinction impossible to miss.
3. 03Laid out the funding structure honestly, including the condition that no bonds are issued and no taxes rise if the state grant does not come through.
4. 04Documented the district's record, the 1996 high school bond paid off and the 2018 override delivered and renewed in 2023, because that is the argument a skeptical voter actually weighs.

Outcome

A campaign site that hands voters the arithmetic instead of asking them to take it on faith, and answers the cost question before the opposition can frame it.

What I did

1. 01Built an auto-syncing feed that aggregates events across venues and primary sellers, with no manual upkeep once running.
2. 02Designed the filtering around how people actually decide, not around how the source data is shaped: what night am I free, how far will I drive, what am I willing to pay.
3. 03Linked every listing straight to the primary seller, so the app routes demand rather than intercepting it.

Outcome

A live, public utility that stays current on its own.

What I did

1. 01Structured the trip as a sequence of legs with their own logistics, so a change in one place propagates visibly instead of quietly.
2. 02Designed for reading on a phone mid-transit, where the plan actually gets consulted.

Outcome

One authoritative plan, legible at a glance, that survives contact with reality.

What I did

1. 01Built and run the brand end to end, from the site through to a 7–14 day delivery commitment.
2. 02Specialized in on-base access and community rates for military families.

Outcome

An operating business that sharpens the commercial instincts the analytics work depends on.