Dylan Walls / Dental operations
Portfolio indexAI systemsPortfolio 02 · dental operations

Flagship case study / AI implementation

Make the dental day legible before you automate it.

A calm, recruiter-facing walkthrough of how I translate a messy dental-office workflow into a bounded operating picture: queue the signal, preserve the boundary, and hand the decision to a human.

Project
Dental Office Operations Control Room
My ownership
Workflow framing, UI projection, evidence boundaries, QA
Recruiter lane
AI implementation · solutions engineering · platform operations
Stack surface
TypeScript source behaviors · static HTML/CSS/JS case study

01 / brief

A control room, not a dental marketing page.

The design goal is not to make the practice look busy. It is to make operational judgment visible: what is known, what needs a person, and what the system must refuse to imply.

01 / PROBLEM

Signals are distributed.

Appointments, intake completion, resource conflicts, clinical follow-up, and billing readiness are different workflows. A useful surface should connect the handoff without pretending they share one source of truth.

Verified behavior
02 / OUTCOME

Reviewable next action.

The walkthrough moves one synthetic case from an operations queue to an intake signal, then to a clinical signal and a human review handoff. Every transition is visible and reversible.

Observed prototype
03 / GUARDRAIL

Boundaries stay on screen.

No PHI, claims, payments, signatures, persistence, or production outcome is hidden behind the interaction. The page is a browser-local read-model demonstration.

Synthetic only

02 / critical path

Queue the signal. Keep the human in the loop.

Start with the work queue, inspect the synthetic appointment and intake context, surface the clinical follow-up, and route a review request. This is a deterministic local fixture: no network, no save, no submission.

Dental Office Operations / local walkthroughLocal fixture ready · synthetic only
Nothing is saved or submitted from this surface.
State lab

03 / architecture & evidence

Make the seams inspectable.

The important architecture decision is not “where can AI go?” It is “where does a signal stop being a projection and require an authorized human or connected service?”

Observed prototype Data-flow map for the proposed control-room path. Solid arrows describe the browser-local demonstration; dashed arrows are follow-up seams.
Dental office operations data flow Browser UI reads a bounded projection, passes through an API and workflow boundary, reaches proposed persistence, and stops at a human review seam. BROWSER UICommand centerQueue · states · reset API / DATA BOUNDARYBounded projectionValidated read model WORKFLOW LOGICSignal classificationHandoff / no auto-write PERSISTENCE BOUNDARYAudit + recordsPROPOSED · not connected HUMAN REVIEWClinical / ops ownerApprove · resolve · amend readclassifyfuture writeaudit trailreview request
Observed prototypeProposed seam

Boundary 01

Read model first

The source implementation uses a typed presentation projection with explicit safe states. That gives the UI a truthful loading, empty, or error posture before any clinical or financial value is shown.

Boundary 02

Human review is a state

A queue item can be classified and routed without becoming a diagnosis, signature, claim, payment, or scheduling mutation. The demo ends at the handoff.

Boundary 03

Persistence is proposed

Production work would add authorization, audit, idempotency, append-oriented clinical history, and recovery. Those are shown as a seam, not claimed as delivered here.

04 / narrative

The decisions behind the surface.

A recruiter should be able to scan the story quickly and still find enough technical detail to ask a useful follow-up question.

Problem

Operational ambiguity has a cost.

Staff do not need another dashboard full of counters. They need to know which signal is urgent, what context is safe to show, and who owns the next decision.

Constraints

Trust beats theater.

  • Synthetic records only; no PHI or copied production data.
  • No connected claims, payments, signatures, persistence, or live outcomes.
  • Mobile-safe, keyboard-accessible, and calm under reduced motion.

Approach

Projection → signal → handoff.

I kept the browser artifact small and deterministic while preserving the source behavior vocabulary: schedule/resource status, intake completion, clinical note follow-up, and honest not-connected states.

Key decisions

Make “not connected” useful.

Unavailable data is a product state, not a blank card. Planned persistence is visually separated from observed prototype behavior so a recruiter can see where implementation work remains.

05 / evidence & QA

Inspectable, not inflated.

The source project has its own validation suite. This case study records what that evidence supports and what this static route adds as a review surface.

Verified source behaviors

The source command-center projection renders deterministic synthetic practice context, appointment/resource statuses, attention items, unsigned-note follow-up, exact money fields, and claim-readiness labels. Its tests also cover safe states, escaping, navigation, and responsive contracts.

Source evidence Reference: integration-command-center worktree at commit 8c47c3d.

Observed prototype behavior on this route

This page's walkthrough is intentionally browser-local. The four-step state machine, reset behavior, empty state, unavailable-source state, and loading state are implemented in inline JavaScript with no network reads and no storage writes.

Route evidence Verify by opening the walkthrough, advancing through the handoff, previewing state lab buttons, and resetting.

What is explicitly not claimed

This is not a HIPAA certification, production deployment, clinical decision support system, payment processor, clearinghouse connection, scheduling system, or measured customer outcome. It is a sanitized portfolio artifact.

Proposed follow-up Connect only after authorization, audit, persistence, test, and operational recovery gates are defined.

06 / claim ledger

Every claim has a lane.

The ledger keeps verified source behavior, synthetic demo fixture, observed prototype behavior, and proposed follow-up separate.

Case study claim and provenance ledger
ClaimClassificationEvidence / treatment

Command-center projection surfaces operational, clinical follow-up, and claim-readiness signals.

Verified source behavior

Source renderer and tests; represented here with sanitized fixture labels.

Walkthrough can route a signal to a human review seam without writing a record.

Observed prototype

Deterministic local state machine in this route; resettable and no network.

Displayed people, appointment labels, counts, and statuses are real operational outcomes.

Not claimed · synthetic only

All records are fictional fixture values and visibly marked as such.

Authorization, audit, persistence, and recovery seams should be connected next.

Proposed follow-up

Architecture map shows dashed, disconnected boundaries; no production implementation is implied.

07 / next iteration

The next build earns its writes.

The natural next step is not more visual polish. It is to make each proposed seam testable without losing the calm operator experience.

  1. Connect an authorized read adapter.Define tenant/location scope, bounded payloads, source freshness, and failure behavior before adding live data.
  2. Make review and audit append-oriented.Record who acknowledged, resolved, amended, or rejected a signal; preserve clinical history instead of overwriting it.
  3. Exercise recovery and authorization.Test stale updates, duplicate delivery, unavailable dependencies, and cross-location access before a pilot.