I design AI systems against the friction points of real operations, informed by a career across sales, technical sales, and field application engineering in solar and energy storage, from C-suite to field installer, five-figure to seven-figure deals. I'm not a trained engineer or programmer; that turned out to be the angle that produced the work below. The portfolio includes a governed support-knowledge console deployed in daily use inside a multinational manufacturer's support organization, a complete beta-program architecture for a hardware NPI program, a document-intelligence engine in production with paying clients, a 74-module measurement engine with 43 USPTO provisional filings, a live consumer reference app, a cross-domain falsification that closed on null, an inverted-architecture solar vertical that reached the same data-access ceiling that limits the broader engine, an EU AI Act reference architecture killed within a day of a self-commissioned competitive review, and an operating California advisory practice where AI is auditable production infrastructure under a published Terms of Service and attorney-reviewed engagement agreement.
The work below runs in two layers. The first is deployed operational tooling: a governed support-knowledge console in daily use inside a support organization, a beta-program architecture running a hardware program, and a document-intelligence engine with paying clients. The second is the measurement substrate underneath: primitives, attestation interfaces, and open standards, tested for whether they transfer across domains. Each layer feeds the other; the primitives earned their claims by shipping against them.
The portfolio is structured around falsifiability. Each entry declares a bar. Several clear it; two close cleanly on failing to clear it; one closes against a structural ceiling that limits the broader engine itself. A commercially credible design can still fail as a business, a sound architecture can still lose to a free incumbent, and clearing synthetic-substrate tests says nothing about real-data validation. The closures are featured rather than buried, because a portfolio that only shows wins demonstrates selection, and I want it to demonstrate judgment.
A career across sales, technical sales, and field application engineering, C-suite to field installer, five-figure to seven-figure deals. That built the cognitive substrate the AI work draws from: pattern recognition across modalities, liability instinct, cross-system reasoning, and the habit of operating in the gap between what a customer says and what's actually wrong. Operating exposure inside five multinationals with markedly different norms around risk, hierarchy, and contract precision. Tigo Energy through IPO preparation, involved in some of the decision-making and preparation during the run-up, watching the legal, compliance, and liability apparatus get built around an operating business in real time.
Broad architecture funds vertical execution. PIE is sensor-agnostic, domain-agnostic, worldwide-applicable, and stands as a substrate. Every commercial extraction came from narrowing: FretMind, WindPIE, SolPIE, Cardinal, The Installer's View, The Installer's Lens. Narrow versions moved faster and reached defensible architectural conclusions the broad version never could. Both layers are necessary, because broad investment generates the substrate that vertical execution consumes. All six extractions inherit the same structural ceiling: synthetic substrates and public data can take work to defensible architectural conclusions, but real predictive validation requires institutional data access that solo investigation cannot obtain. Applied to AI: frontier capability is broad; products that survive pick a vertical and execute hard while capability stays general.
AI was used throughout as a structured tutor and architect rather than a code generator. Frontier LLMs walked me through unfamiliar territory step by step while I executed the actual work, with active pushback when something didn't fit. The work shipped at the same speed it would have anyway, and the skill compounded instead of being outsourced.
I trust my intuition because it has been trained against consequences in the field; whether the work below earns that trust is the reader's call.
A single-file offline support console in daily use by a manufacturer's customer support and engineering team. The interface is a compiled artifact; the asset underneath is a governed, citation-complete, machine-readable knowledge corpus.
All content lives as structured data: entities, procedures, reference facts, configurations, Q&A entries, cross-links, and search aliases. A deterministic Python assembler compiles that data into a dependency-free HTML file that runs from a file share with no network calls. The same input always produces the same md5, and every build passes a battery of checks before it ships. An authored concept and intent router sits in front of a scored literal matcher with alias mapping and typo tolerance; the build-side audit tooling and the shipped JavaScript are parity-tested against each other, so the test harness and the product provably rank results the same way.
Several hundred per-fact citation bindings, each resolving to its source document; builds fail hard on any stale or unknown reference. Attribution carried as data: every fact is tagged by the class of authority behind it, with a precedence model governing conflicts between source types. Gated ingestion with regression baselines covering search relevance, link integrity, reachability, and terminology. Coverage gaps are declared instead of hidden.
The deterministic search router doubles as the retriever, and the model synthesizes answers under a system prompt built from the tool’s own trust rules: answer only from the provided material, decline anything outside it, and escalate when the material is thin. In its first measured evaluation it scored a 1.0 retrieval hit rate and a 0.0 false-confidence rate on the curated battery; every out-of-scope and wrong-product probe was declined instead of answered. The corpus ships with its own evaluation infrastructure, a purpose-built bank of more than a thousand field-realistic questions tagged with expected answers.
The analytical engine underneath the advisory practice, in production and powering paid client deliverables: a Python-based multi-source verification platform that automates a labor-intensive technical review workflow.
California homeowners evaluating a rooftop solar decision routinely receive economic projections built on outdated tariff assumptions, despite the state's transition to NEM 3.0 / NBT in April 2023, under which export credits fell substantially. Independent technical verification has historically been slow and expensive relative to the size of the decision. The Lens automates that verification workflow at a fraction of the cost and turnaround.
Single-user Streamlit UI wrapping a Python backend that parses unstructured intake (utility bills via vision-model OCR, Green Button XML, equipment photos, proposal documents), orchestrates parallelized API queries across 16 authoritative public data sources, runs a 10-lens analytical framework, and generates client deliverables via template-driven PDF rendering with full audit trail. Sources span solar resource modeling (NREL NSRDB, PVWatts, PySAM), roof geometry (Google Solar API), equipment validation (CEC), public regulatory and legal record sources, environmental context (EPA AirNow, CAL FIRE, CPUC PSPS), and utility-specific NBT rate schedules.
Sole architect and product owner, directing full implementation through an LLM coding agent under written standing instructions, with human approval gating each merge; every architectural decision, data model, and release passed a structured multi-role design review before build. Safety and integrity are structural: fail-closed vision-based PII redaction on every document format before any data leaves the machine, a six-gate intake validation system, copy-fidelity canaries that fail the build if rendered prose drifts from approved text, and a principal-verification seam that makes it mechanically impossible for machine-drafted judgments to reach a client without human approval. Verified by a 1,200+ test suite with typed contracts and in-code API budget guards. Local-first storage (SQLite + per-engagement file structure) for audit defensibility. Fifth extraction of the PIE primitives: baseline-measurement intelligence applied to system-level economic verification.
A small independent California advisory practice, operated solo, and the working context in which The Installer's Lens runs. Included here for the operating model rather than the business: a professional practice where AI is production infrastructure under documented governance.
Hand-coded on a static site generator with a custom design system, version-controlled and continuously deployed, with no theme, page builder, or plugins and no visitor tracking. Operations run against one canonical, append-only decision record with a read-before-write conflict protocol. Engagement terms were drafted under a documented AI-disclosure framework and routed for attorney review, and the practice carries professional liability and cyber coverage. Entity, banking, insurance, and trademark work were stood up independently.
The complete documentation and process architecture for a hardware beta program, covering site qualification through closeout, built inside a manufacturer's field applications organization.
A field-scoped SOP with on-site capture requirements, a pass/fail site qualification system, a program workbook serving as system of record, structured Voice of Customer instruments, a per-site process record, and an illustrated field capture guide. Legacy tracking modernized from a sprawling spreadsheet to a focused field set, with static pivot reporting replaced by live formulas.
All documents produced through AI-assisted programmatic generation (Node.js docx, Python openpyxl, vector illustration pipelines), with iterative direction, domain review, and final technical judgment retained by the principal. Weeks of drafting compressed into days, with cross-document alignment enforced by the pipeline itself, including identical checklist content between the SOP appendix and the live workbook tool.
The consumer reference app that proved PIE works in a live capture loop. The meta-artifact on which AI collaboration patterns produced novel work versus runaway scope.
Browser-based: Web Audio API for real-time pitch and rhythm, MediaPipe Hands for body mechanics, Basic Pitch (Spotify) for audio-to-MIDI. Five stateful coaching personas route to different LLM prompting and feedback styles. Welford-based individual baselines with z-score classification against the player's own history. DALD monitors divergence between AI-claimed session quality and the player's independently-measured improvement trajectory. VTACA detects breath-holding patterns during cognitive load. PBITE gates interventions on practice-quality signal.
Persona system as a structural product decision: each persona routes to different prompting and recommendation libraries. Module triage against consumer use: a subset kept, a subset simplified, the remainder removed or retained as reference only. Shipped the hand-written browser port first rather than waiting for full-engine integration, prioritizing real sessions over architectural purity.
Live play sessions with real audio capture, verifying breath-hold detection and intervention gating worked in production rather than in simulation. Surfaced a mic-clipping problem that traced to input sensitivity rather than engine failure. Forced the distinction between "commercially credible" and "commercially viable" and admitted FretMind achieved the first but not the second. None of PIE's core functional claims (quality score validity, PBITE gating, DALD, BAIV, CAAD, AICV) have been empirically validated on real deployment data.
The substrate the deployed work draws on, plus the projects that closed. Summaries only. The two drafts below (IBTR / TRIL and DALD) are exactly that: unpublished working drafts, not peer-reviewed papers and not currently being pursued. They would need empirical validation I have not done before they could be submitted anywhere. They are here because the reasoning is mine and the bars are declared, not because they are finished.
A sensor-agnostic measurement layer that scores any system's real-time state against its own historical baseline. Humans, hardware, AI systems, field-deployed sensors. Never against a population.
The substrate underneath the deployed work: individual-baseline measurement with an open-standard interface, sensor-agnostic and domain-agnostic by construction. 43 USPTO provisional applications filed April 2026. Every commercial extraction below came from narrowing it. Its core functional claims have not been empirically validated on real deployment data, and the ceiling is data access rather than architecture.
The foundational paper underneath the seven other projects in this portfolio: population reference is categorically wrong for individual-divergence questions, and one self-referential measurement substrate operates across domains as different as music, surgery, AI alignment, industrial machinery, and seismic monitoring.
An unpublished working draft arguing that individual-baseline measurement, rather than population comparison, is the correct primary detection path, with an attestation interface that lets a claim be verified without exposing the underlying data. The cross-domain instantiation is reasoned rather than tested. Not being pursued at present; it would need empirical validation before it could go anywhere.
A narrowed extraction of PIE primitives applied to wind turbine fleet analytics. The hypothesis was that per-turbine individual baselines would beat fleet-mean comparison at surfacing subtle degradation. The data said otherwise.
A cross-domain test of whether the individual-baseline commitment held outside its origin domain. It did not: peer-first detection beat individual-baseline detection by roughly five times on the labeled events. Closed on a bar declared before the work began, and the result directly produced the inverted architecture used in SolPIE.
Built after WindPIE returned null on the original individual-baseline-first thesis. SolPIE inverted the architecture in response: peer-first detection, environmental conditioning instead of output conditioning, variance shift disabled by default. Engine and synthetic substrate built; integration test passes; real-data validation blocked by the same data-access ceiling that limits PIE itself.
The inverted architecture built in response to what the wind work refuted: peer-first detection as the primary path rather than a supplement. Passed synthetic-substrate integration, then reached the same ceiling that limits the broader engine, because real predictive validation requires vendor-grade per-module telemetry that solo investigation cannot obtain.
A reference architecture for EU AI Act compliance infrastructure, scoped as documented specifications plus illustrative TypeScript implementation. Six articles in scope; three explicitly excluded as discipline. Killed within ~24 hours of a self-commissioned competitive ultrareview.
An EU AI Act reference architecture, killed within a day of a self-commissioned competitive review that surfaced a free, permissively licensed incumbent occupying the same position. The kill is the point: the review was commissioned specifically to find the reason not to build, and it found one before the build absorbed real time.
The alignment-specific application of IBTR / TRIL: detecting deceptive alignment in deployed AI systems without requiring model internals access.
An unpublished working draft on detecting behavioral drift in deployed AI systems without access to model internals, on the premise that most parties who need to evaluate a deployed system will never see inside it. It reads the trajectory of the human the system affected rather than the system itself, using per-subject baselines and an online variance estimator. Empirical support is preliminary and single-domain. It does not address sudden catastrophic failure, only slow drift, and it would require validation I have not performed before it could be submitted. Not currently being pursued.
I'm the profile that doesn't show up in a standard candidate pipeline. A senior operating record across sales, technical sales, and field application engineering, the last several months of intensive AI building from a standing start, a support-knowledge console deployed in daily use inside a manufacturer's support organization, a document-intelligence engine in production with paying clients, a 74-module measurement engine with 43 USPTO provisional filings, two projects killed cleanly on bars I set in advance, an inverted-architecture solar vertical (SolPIE) built in direct response to what the wind work refuted, and an operating California advisory practice where AI is auditable production infrastructure under a published Terms of Service and attorney-reviewed engagement agreement.
Not a trained engineer or programmer. That's the point. The work above is what someone with my background builds when AI removes the bottleneck that would otherwise have required hiring an engineering team. And the discipline to instrument, falsify, and kill the work cleanly was already there, built by a career of carrying responsibility for outcomes in front of customers, lawyers, and regulators.
Primary target: Product Support / Customer Support Engineering Management roles at AI companies, where deep customer-support operations expertise and substantive AI literacy combine. The lateral move from solar industry customer support into AI customer support is deliberate: same function, adjacent domain, with demonstrated rapid AI adoption.
Sales-led roles: Enterprise Sales, Strategic Account Executive, Business Development, and Director-level commercial roles at AI and AI-infrastructure companies. A senior commercial track record (territory growth from $250K to $15M+, first to $1M quarter, first to $1M month, Director of Sales) paired with the AI portfolio above is a rare combination in the AI hiring pool.
Technical and customer-facing roles: Solutions Engineering, Forward-Deployed Engineering, Technical Sales Engineering, and Technical Product Manager at AI and AI-infrastructure companies. Also open to founding Solutions Architect or founding Product roles at AI startups under twenty people, where the seat pairs deep AI understanding with operational instinct and customer-facing credibility, and engineers own the implementation.
Generic version. Role-specific resume variants are submitted directly to each application.