Technology / Systems

Product direction,
proved in systems.

I turn ambiguous problems into inspectable products: defining the decision model, directing AI-assisted implementation and testing what actually holds up.

AI product direction / creative technology / validation systems

05Evidence-led case studies
1Published falsification — claims here are tested, not promised
ExplicitAuthorship and outcome status
Evidence standard

Claims need
receipts.

Each principal case separates what exists, what I directed, what AI agents implemented and what has not yet been validated outside the build.

01

Exact Role

Product decisions / system constraints / interaction direction / acceptance criteria / review responsibility

02

Build Evidence

Architecture / dated checkpoints / test records / generated outputs / known failure cases

03

Clear Attribution

Johan-led direction / AI-assisted implementation / upstream dependencies / third-party boundaries

04

Outcome Status

Local validation / user evidence / business evidence / what remains before a production claim

02 / AI Security
Interactive prototype

Gatewarden

A local security gateway that gives people visibility and control over the tools used by AI agents.

Role
Product direction / security workflow owner
Contribution
Threat model, product definition, interface direction, acceptance criteria
Attribution
Johan-led direction; substantial implementation by Claude agents
Stack
TypeScript, Node.js, Electron, MCP
Interactive product prototype / mock dataOnboarding and exposure scan
Product views / Current prototype02 additional screens
Protection dashboardDecisions / alerts
Evidence-report prototypeRepresentative output

Problem

AI clients can connect to tools that change definitions, expose secrets or introduce unreviewed actions. Existing controls are often too technical for an individual operator.

System

A transparent local proxy inspects every message, pins approved tool definitions, gates new connections and records signed, privacy-aware decisions.

Product decision

One engine has two interfaces: a CLI for technical inspection and a restrained menu-bar application for everyday control.

Evidence / recorded buildNot production deployed or independently audited
101Engine tests in recorded build
77Application tests in recorded build
0npm audit vulnerabilities reported
0Semgrep findings reported
  • ImplementedProxy defenses, exposure scan, audit logging, CLI and Electron interface
  • ImplementedEngine-to-app connection, onboarding, configuration backup and restore
  • OutstandingLive Claude Desktop verification, signing, notarization and production licensing
Build evidenceAvailableArchitecture, source modules and adversarial tests
User evidenceNext milestoneFirst external operator session; none claimed yet
Business evidenceNext milestoneSigned public beta, then first deployment; no revenue claimed yet
Inspect Gatewarden build receipt
03 / Product Systems
Built locally / pre-launch

Wandrize

A source-linked prototype designed to turn complex visa requirements into clearer preliminary guidance.

Role
Founder / product and UX direction
Contribution
Rule model, customer journey, verification workflow, acceptance review
Attribution
Johan-led product; AI-assisted research, design and implementation
Stack
Static web, JSON modules, serverless APIs, Stripe test flow
Working local interface / pre-launch productRules translated into a guided decision
Product view / Generated experience01 additional screen
AI answer hubProof-aware output

Problem

Visa information is fragmented, frequently changing and difficult to translate into a personal decision without exposing sensitive user data.

System

A reusable profile is evaluated client-side against structured country and visa modules, with blockers, confidence notes and source records retained alongside each rule.

Product decision

Each country is a data module rather than a new website, allowing one interface and verification workflow to scale across markets.

Evidence / current buildBuilt components; not commercially validated
25Country configuration modules found
74Generated country and audit pages
54→84Internal AI-search readiness score, before→after fixes
LocalEligibility logic runs in the browser
  • BuiltEligibility experience, modular country structure and serverless payment functions
  • DocumentedSource policy, verification logs and shared rules schema
  • OutstandingDeployment validation, real customer testing and commercial proof
Build evidenceAvailableRule engine, payment flow, monitoring and country modules
User evidenceNear zeroNo reliable task-completion or trust data yet
Business evidenceZero revenueRecorded in the July external-validation brief
Inspect Wandrize build receipt
01 / AI Operations
In development

SignalLayer

A local AI-search and SEO operations system that turns a website scan into proof-gated recommendations, staging fixes and repeatable monitoring.

Role
Product direction / validation-system owner
Contribution
Product model, evidence gates, approval logic, workflow and interface review
Attribution
Johan-led product; substantial implementation by AI agents
Stack
Python, local web app, structured data, staged publishing packages
Working local interface / workspace lens deskNext action, evidence and approval state
Product views / Current local build02 additional screens
Validation labEvidence gates
Design conceptNot product evidence

Problem

AI-search and SEO tools often stop at a score, while automated publishing can make unverified claims or apply risky changes without a usable review path.

System

A local control plane captures a site, scores readiness, generates staged fixes, records evidence and separates safe recommendations from actions requiring proof, credentials or approval.

Product decision

Every recommendation carries status, evidence and a next action. Production publishing and paid provider calls remain explicitly gated.

Evidence / dated checkpointRecorded 2026-08-02; not independently rerun
827Tests in the recorded checkpoint
170/170Routes in recorded validation
16/16Rendered QA captures recorded
258Files in export-safety check
  • ImplementedSite capture, local readiness audit, action queue, reports, rollback records and publishing packages
  • RecordedSafe local audit completed against this portfolio preview; the live Squarespace site remained untouched
  • OutstandingLive prompt monitoring, Search Console imports, provider routing and hosted customer workspace
Build evidenceStrongestCode, tests, QA captures and operational records
User evidenceInternal onlyNo external customer workflow study yet
Business evidenceNext milestoneSigned public beta, then first deployment; no revenue claimed yet
Inspect SignalLayer build receipt
04 / Applied Research
Research build

Opening Range
Strategy Lab

A modular futures-strategy research system designed to reject attractive backtests when they fail costs, holdouts, walk-forward tests or platform checks.

Role
Research and validation-system direction
Contribution
Experiment design, risk rules, falsification gates and result review
Attribution
Johan-led research; AI-assisted implementation; upstream strategies attributed
Stack
Python, pandas, NinjaScript / C#
Research pipeline / recorded buildNot live performance
Candidate / MNQ inside-barFalsify
before deploy.
  1. 01Cost-aware backtestFees / slippage / session logic
  2. 02Walk-forwardOut-of-sample windows
  3. 03Placebo testsChallenge false edge
  4. 04Frozen holdoutKeep final data untouched
  5. 05Platform parityNinjaTrader verification remains
Recorded falsification result
4.56Training profit factor
0.95Out-of-sample profit factor
RejectPromotion decision

The edge disappeared out of sample. The configuration was classified as curve-fitted rather than promoted.

Validation architecture / research harness / explicit rejection gates

Problem

Trading ideas can look persuasive in a naive backtest while failing once realistic costs, new periods, platform behavior and risk limits are introduced.

System

A modular harness separates data loading, strategy logic, metrics, prop-firm simulation and diagnostics, then moves candidates through falsification gates before a NinjaScript port.

Research decision

Rejected configurations and risk-cap breaches remain in the record. A promising test is evidence for the next test, not proof of a deployable trading system.

Evidence / recorded rejectionResearch evidence, not performance marketing
4.56Training profit factor recorded
0.95Out-of-sample profit factor recorded
5Named validation gates
RejectedCurve-fitted configuration
  • ImplementedCost-aware engine, diagnostics, trade-level exports, stress checks and NinjaScript candidates
  • RecordedPassing and rejected configurations, degradation checks and explicit promotion decisions
  • OutstandingNinjaTrader parity, paper trading, live risk verification and independent result review
Build evidenceAvailablePython harness, smoke tests, trades and result summaries
Operating evidenceUser-maintainedNot yet corroborated by redacted broker exports
Performance claimNoneNo expected-return or deployable-edge claim
Inspect Strategy Lab build receipt
05 / Creative Technology
Validated locally

JH Effects

A working image-effects tool that turns photographic direction into repeatable, adjustable processing without removing the final creative decision.

Role
Creative director / workflow owner
Contribution
Effect direction, workflow design, acceptance criteria and local output review
Attribution
Built for Johan with AI-assisted implementation; locally executed and reviewed
Stack
Python, Pillow, NumPy, SciPy, Tkinter
Recorded local run / 2026-08-09Actual source and generated output
SourceOriginal photograph
Generated outputWorn Polaroid / matte warm

Problem

Creative references are difficult to translate into repeatable production steps, especially when variations, borders, textures and double exposures need to remain adjustable.

System

Five processing modes combine modular borders, tonal treatments, textures, double exposure and stacked pipelines through CLI and interactive workflows.

Creative decision

The tool generates controlled variation while keeping image choice, intensity, sequencing and final acceptance with the photographer.

Evidence / local executionOne run verified during this portfolio audit
4,279Lines in the principal Python tool
5Documented processing modes
1Successful local output run
0External user claims
  • ExecutedSource loaded, worn-Polaroid border, matte-warm process, grain and dust applied successfully
  • VisibleThe source and exact generated output are shown above at full resolution
  • OutstandingRepeatable automated tests, packaged distribution and external workflow testing
Build evidenceAvailableSource tool, packaged launcher and generated files
User evidenceOwner-testedLocal run verified; no external study
Business evidenceNext milestonePackage signature grades as a preset pack; no revenue claimed yet
Inspect JH Effects run receipt
Experiments / in development

Technical workbench

Smaller systems that show range, technical curiosity and practical workflow design. These are presented as experiments, not finished products.

E.01

Inventory Eye

A persistent opportunity-research workflow that records existence checks, ranked ledgers, rejected ideas and corrections between scans.

Working workflow21 scans recorded; not a standalone product
E.02

Wrapsheet

An interactive production-estimating prototype for photography and motion, including dual currencies, licensing logic and client-facing worksheets.

Interactive prototypeInterface works; signing, payment and AI drafting are demonstrations
E.03

AuditIQ

A Next.js audit prototype exploring parallel content, conversion, SEO, positioning and growth analysis with structured report orchestration.

PrototypeArchitecture exists; no tests, deployment or users yet
E.04

SecondBrain agent workflows

A local research-vault protocol for ingestion, claim verification, cross-linking, daily briefs and source-grounded query responses.

Early developmentWorkflow documented; source collection is still lightly populated
E.05

Photoshop automation

Readable JSX tools for repeatable glow, brush setup and light-shaping operations while preserving manual creative judgment.

ExperimentScripts written; live-document testing remains
E.06

Market Regime Briefing

A scheduled research workflow for specialist market analyses, stale-data warnings and explicit fallback rules.

Early experimentScheduler and prompts exist; completed recurring reports remain unverified
How I work

Creative judgment.
Technical evidence.

01

Make the system legible

Reduce technical complexity into an interface, workflow and decision model people can understand.

02

Use AI transparently

Direct agents as implementation collaborators, then state the boundary between ownership, direction and generated code.

03

Show what is proven

Separate working evidence from product ambition, and make unfinished risks visible rather than dressing them as traction.

AI product direction / creative technology / validation systems

Build it.
Then prove it.

johanvonevil@gmail.com
System dossier

Scope

    Proof

      Next validation

        Build disclosure