Kyle Kilmetis · Engineering portfolio

Engineering at architect altitude.

# positioning

Cloud infrastructure, geometry, knowledge systems and software that does real work. Open an exhibit. Change an input. Read a piece of the source.

Selected work / Working exhibits

Less promise.
More proof.

Six things I am building, each with something you can operate. Source inspected September 2026; development status is not a launch claim. No client or customer data appears here.

Claims & resolution · Archetype 03 · Resolve

Prove the facts once.

An evidence-driven claim record for restoration and insurance work. Documents, inspections, scope, estimates and approvals live in one connected record, so the same facts are not proven again at every step.

Lifecycle, evidence integrity and scope gate implemented · pre-pilot

Try the reconciliation step: estimate lines are matched against the approved scope, and every difference needs a recorded decision before a proposal can be drafted. Synthetic line items; no claim data.

The archetype behind it →

LIFECYCLE · STEP 11 OF 17 · ESTIMATE REVIEW

LineScopeEstimateDecision
Water extraction, 480 sq ftMatched$960$960No decision needed
Remove engineered flooring, 40 sq ftMatched$520$520No decision needed
Drying equipment, 3 daysVariance$1,140$760
Ceiling drywall, 120 sq ftMissing from estimate$1,450—
Antimicrobial application, 480 sq ftNot in scope—$384

Gate closed. 3 differences need a recorded decision before a proposal can be drafted.

Demonstration · synthetic line items · logic mirrors the application’s reconciliation rule

Functional overview

  1. 01

    Intake

    Loss event, property, customer

  2. 02

    Inspect

    Areas, observations, evidence

  3. 03

    Scope gate

    Verified evidence required

  4. 04

    Reconcile

    Every difference decided

  5. 05

    Propose

    Versioned, second approver

CLIENTSField appoffline-first PWAOffice webreview · approveCRM (optional)adapter onlyNext.jstyped contractsFastAPI · domainLifecycle (28 states)Evidence integrityScope gateReconciliationProposals + approvalAudit trailAI gatewayadvisory only · redactedinput hashed · human review recordedSTORESPostgreSQLrow-level securityEvidenceSHA-256, immutableOutboxqueue, retriesA loss event is not a claim. Scope is not an estimate.A recommendation is not an approval.Versions are append-only; each points to its predecessor.
Logical architecture, drawn from the application’s architecture notes · pattern level
Inspect the actual source excerpt — Allowed lifecycle transitions · Python

The lifecycle is enforced on the server: 28 states with forward, backward, on-hold and exception moves. Backward moves and exceptions require a reason, and every transition is audited. AI output is advisory only, and personal fields are redacted before any model call.

    if current in EXCEPTION_STATUSES:
        return frozenset({LeadStatus.CLOSED})
    if current is LeadStatus.ON_HOLD:
        allowed = {LeadStatus.CLOSED}
        if held_from is not None:
            allowed.add(held_from)
        return frozenset(allowed)

    allowed = set(EXCEPTION_STATUSES)
    if current in PRIMARY_PATH:
        index = PRIMARY_PATH.index(current)
        if index + 1 < len(PRIMARY_PATH):
            allowed.add(PRIMARY_PATH[index + 1])
    allowed.update(_FORWARD_EXTRA.get(current, frozenset()))
    backward = _backward_target(current)
    if backward is not None:
        allowed.add(backward)
    if current in _HOLD_ELIGIBLE:
        allowed.add(LeadStatus.ON_HOLD)
    return frozenset(allowed)

Selected source excerpt · inspected September 2026 · no application data

The earlier arc

Before the cloud, the pitch.

A systems-under-pressure story most cloud engineers cannot tell, and the reason the reliability instinct came before the title.

  1. Systems under pressure

    Video-assistant-referee replay systems deployed across Major League Soccer stadiums: 150+ production servers, operated live for a Cup final and two FIFA tournaments. Match day does not have a maintenance window.

  2. The data side

    Sales engineering at Varonis: data security for enterprises that could not tell you who could read what. Translating a threat model into a deployment a customer would actually run.

  3. On my own

    Independent architecture consulting. Small teams, real constraints, systems that had to work the first time because there was no second budget.

One page, when you need it.

The résumé is on request until the PDF is published here.