Thomas CareySCIENCE · TEACHING · SOFTWARE · PUBLIC SERVICE

SOFTWARE

Software I built.

I use AI coding assistants to turn problems from public service and training into working software. I define what the tool must do, try it in context, and revise it when it falls short.

Working applications

Project Arete interface

Project Arete

Organize policy review

Keep drafts, comparisons, discussion stages, and decisions within reach.

Project notes Open app
Project Excelsior interface

Project Excelsior

See the whole training record

Explore training trends, workout planning, and a saved example of AI coaching.

Project notes Open app
Work Session Visualizer interface

Work Session Visualizer

Run a meeting

Arrange a discussion, manage the clock, and export a Word summary.

Project notes Open app

SMALLER WORKFLOWS

Less time on the paperwork.

Meeting transcripts & chapter markers

Find the useful part of a long meeting.

A meeting recording is much easier to use when you can search its words and jump to the discussion you need. My transcription workflow produces a transcript and draft chapter markers: named timestamps for sections of the recording. Board-office staff can review them for publication.

The finished transcript makes the meeting searchable, and the chapter markers give viewers useful entry points into the recording.

A long meeting recording3:12:00
Transcript + chapter markers
Opening & agenda
Budget discussion
Policy review
1:36:42

“Let’s turn to the proposed changes and the questions raised during review…”

Illustrative output · find the passage, jump to the discussion.

Batch policy redlines

Compare a whole batch of policies.

Upload current and proposed policies as individual files, mixed PDF/DOCX batches, or ZIPs. The tool identifies each policy by its code, matches the versions, and can split a combined series PDF into individual policies. It also recognizes new and repealed policies.

A single download contains PDF redlines that mark added and deleted wording, plus an index showing which source files were matched and how each policy was processed.

Current + proposed policy batches
PDFDOCXZIPCombined series PDF
Identify policies · match versions
MATCHED BY POLICY CODE

Policy AB · proposed changes

The committee will review the policy every three two years.

One download for the batchPDF redlines + source-matching index
Illustrative output · from mixed source files to matched comparisons.

Policy Explorer · completed pilot

Classify policies for review.

I built a personal policy-analysis web app to sort policies into Board work, staff work, or possible repeal. I designed a four-agent workflow with separate prompts for statutory requirements, operational risk, classification, and adversarial review. When the critic flags an error, the classifier revises its answer and the critic checks it again.

The app also exported recommendations, explanations, and auditor flags to Word and a spreadsheet. In one saved example, the auditor caught a claim that a statute required something it only encouraged. That is why I designed the agents to challenge and correct one another before I reviewed their recommendations.

District policies + Colorado statutes

FOUR AI AGENTS · SEPARATE PROMPTS

  1. Statutory analyst

    Checks whether the statute requires a policy.

  2. Operational risk analyst

    If not mandated, checks whether it is still necessary.

  3. Governance classifier

    Recommends Board work, Staff work, or Repeal.

  4. Adversarial critic

    Checks the classification and its statutory claims.

Flagged? Back to Agent 3.

The classifier revises; the critic checks again.

Recommendations for human reviewWord summary + spreadsheet
Core-policy workflow · flagged results get a second pass before export.

Mileage reporting

A mileage report, maps included.

Starting with the mileage spreadsheet, the workflow sorts the trips, looks up driving distances where route documentation is required, and retrieves clean route images through Google Maps APIs.

It assembles a formatted PDF with the trip table and labeled route maps, plus an editable spreadsheet. The maps arrive in the report, already paired with the trips they document.

Mileage spreadsheet + Google Maps
Trip recordsDriving distancesRoute images
Assemble the reimbursement PDF
MILEAGE REPORT · PDF

The trip and its route, together

School visit12.4 miles
District officeSchool
Illustrative layout and schematic map · the workflow inserts Google Maps route images.

BUILDING WITH AI

You can just do things.

AI helps me learn unfamiliar material, write and revise code, and investigate failures. I bring the context and decide what a useful result would look like. The work is finished when I can use the result for its intended purpose.

For mileage reimbursement, that meant a finished packet with distances, labeled maps, and a formatted PDF. For policy review, it meant accepting the mixed documents staff actually have. These are the habits I want students to practice: choose a purpose, learn what is needed, and test whether the result does the job.

What I would ask students to build