Profile
Twenty years turning environmental science and regulation into working systems, on every side of the field: writing the standards, certifying against them, advising the multinationals that must meet them, and now building the AI-enabled software that codifies them. Has stood up two practices from nothing and owned the quality bar for a third, led pursuit teams that won roughly $2M in engagements, and closed six-figure engagements from personal executive relationships.
Builds AI systems for regulated domains, anywhere the deliverable has to survive an auditor, a regulator, or a board. Sustainability and ESG disclosure is where the proof is live; the discipline transfers to any domain where the source of truth is a document with legal weight. Six systems shipped end to end under a repeatable method: brief, PRD with success and kill criteria, architecture, agentic build with quality control at every step, independent-model verification, and a human accountable for the final assessment.
Model-agnostic by design: Claude Code, an agentic coding tool that reads the codebase, edits files, runs commands, and connects to external tools, drives the build from terminal and IDE, with a second instance overseeing the first; frontier LLMs handle extraction and judging, a different model always verifying the one that did the work; open-source models run locally (Llama via Ollama, Nomic, BERT/SBERT) where data sovereignty requires it. Builds multi-agent pipelines as products, not only as tooling: orchestrator, sub-agents, and verifier with resumable state.
Executive, Commercial, and Delivery Record
- Executive trust. Advised ISEAL, Better Cotton, H&M, Adidas, Nestlé, Amazon, Meta, and the Gold Standard Foundation on value chains and chain of custody, corporate GHG inventories, and biodiversity; delivered Net Zero and disclosure strategy to Fortune 500 and mid-market leadership at Anthesis; recruited by Apex to build a practice from a standing start.
- Revenue. Led pursuit teams at Anthesis that won roughly $2M in engagements; separately closed two six-figure engagements sourced from own executive relationships; currently own go-to-market, pipeline, and client pitching for a new service line at Apex.
- Mobilizing delivery. Teams of 30+, direct reports developed, scope and budget owned, quality and knowledge-management systems built so delivery survives turnover.
- Applied AI, not slideware. Six shipped systems (below), built spec-first with AI agents and frontier LLMs, a different model always verifying the build, governed so models classify and retrieve during construction and are removed from the answer path or fenced behind a trust boundary at runtime.
- Learning agility. Forest ecosystem simulation modeling under climate change to standards authorship to knowledge management & QAQC to verification practice to full-stack AI systems, each ramp self-directed.
AI-Assisted Tools, Apps & Platforms
Built to test how far AI assistance takes one practitioner from published standard to working compliance interface. Requirement registers extracted from source PDFs through a vision-transformer layout and table pipeline: IFRS S2 (1,623 requirements), CSRD (6,016), GRI, SASB, EU Taxonomy, GHG Protocol Land Sector and Removals. Extractions verified by an independent-model grounding check (precision: found / not found / mismatch / interpretation, with counts) before human QA against a set acceptance threshold. SQLite schema, REST API on Cloudflare Workers, dual-token auth, evidence object storage, React front end.
AI-assisted up front; no LLM generates answers to users at runtime.
Governed Answer Retrieval (GAR): every answer traces to the clause it came from and is served verbatim. GAR solves what RAG cannot when technical answers must be 100% accurate: content is human-graded and approved before serving (100% human eval as the release gate), and clause-level grounding makes citation accuracy deterministic. No generation surface in front of the user.
Statute and regulations classified into a structured requirement register. Python content factory, generated JSON, deploy gated on deterministic validation checks (schema and content graders) so a failed check leaves the last good version serving.
DocReview: multi-agent document analysis pipeline Private AI agent pipeline
AI-assisted up front; no LLM generates answers to users at runtime.
Orchestrator with resumable state, corpus manifest, and schema versioned in Git. Seven rubric modes including asymmetric critical review against authoritative references. Every finding carries a support level (dimensional scoring: explicit / inferred / uncertain / missing), never a collapsed numeric score; uncertain or conflicting findings are routed to a verifier pass (scoped LLM-as-judge). Rubric calibrated on two documents before any corpus run.
Second Brain: local-first agentic RAG for technical standards Private Local AI tool
Dual-path retrieval (semantic plus keyword/metadata) with transparency on completeness (the user always knows whether they saw 5 of 5 or 5 of 500) and a keyword-only mode that gives recall by construction when exhaustiveness is required. Eval criteria set in the PRD before build: 100% recall on metadata-filter queries, top-5 semantic relevance ≥80% on a test query set, zero fabricated citations. Trust boundary between deterministic extraction metadata and LLM enrichment; dual-write storage with rollback; consent gate before any outbound call. Delivered from a BMAD PRD with SMART success criteria, kill/pivot triggers, and a traceability matrix.
Outside the domain, to prove the method travels. Full FRS5 rules engine as pure TypeScript with engine/rendering separation; 3D WebGL board; in-browser alpha-beta/PVS opponent with transposition table and quiescence, running in a Web Worker. Evals built against a reference implementation adopted as oracle (gold-standard ground truth): replay of the annotated sample game (recall: every legal move accepted) plus per-position differential checks submitting every generated move to the validator (precision: no over-generation), run as a regression eval suite; deviations from the reference documented and cited.
How I Build with AI
- Spec first. Brief → PRD with success criteria and kill triggers → architecture as the contract the PRD must satisfy → epics and stories → implementation. Method-literate (BMAD, spec-driven development, agentic workflows), not method-dependent.
- Two-model build loop. One Claude Code instance implements; a second oversees, issues prompts, and receives reports; I adjudicate.
- Quality control at every step. Predetermined checks run by an agent after each implementation step, so a failure localises to a stage.
- Evals before code. Eval criteria are written into the PRD before implementation: recall and precision targets, zero fabricated citations, explicit acceptance thresholds. They run as regression eval suites against ground truth (a reference implementation or the source document), so every change is re-verified, and any LLM-as-judge pass is scoped and calibrated before it counts.
- Independent verification of extractions. When AI extracts content that becomes source of truth, a different model checks every sentence against the source and returns found / not found / mismatch / interpretation with counts, a precision signal and a hallucination ceiling, before human review.
- Threshold QA and human accountability. End-of-pipeline acceptance against an explicit threshold set up front; a human makes the final assessment, with tooling routing that human to uncertain cases and a sample of the rest. Earlier R&D used SBERT similarity with a 0.75 threshold to flag likely non-compliance for human assessment, and established where similarity stops being truth.
Experience
Principal, Product Sustainability · Apex Companies
Arkansas, February 2026 to present
Recruited to stand up and lead a product sustainability practice from nothing, owning the service line end to end.
- Own service design, market positioning, go-to-market, and pipeline: build and manage the client pipeline in HubSpot, pitch alongside sales, coordinate outreach, prepare client materials, represent the firm at conferences.
- Lead product sustainability consulting and verification: Product Carbon Footprint, Extended Producer Responsibility, Environmental Product Declarations, life cycle assessment, supplier-specific emission factors, value chain interventions, supplier engagement.
- Build AI-enabled, full-stack tools that codify regulatory logic and institutional knowledge into repeatable delivery.
- Translate regulatory and market analysis into service innovation for clients in a compliance-driven market; partner with marketing to turn technical content into positioning.
Director of Sustainability and Innovation · Sustainable IQ
Arkansas, November 2024 to February 2026
Founded and ran an independent practice delivering carbon accounting and decarbonization strategy, and built the systems behind it.
- Architected and built the multi-tenant ESG compliance platform, the document ingestion pipeline converting regulatory PDFs into structured requirement registers at scale, and the governed retrieval portals above.
- Applied LLMs, retrieval-augmented generation, and BERT-family models to codify regulatory logic into operational tools, governed so models classify and retrieve rather than generate answers.
- Delivered corporate carbon accounting and decarbonization strategy from GHG inventories and Scope 3 value chain analysis.
- Integrated quality and knowledge management systems to build institutional memory that survives turnover.
Associate Director, Corporate Carbon Accounting · Anthesis
Arkansas, October 2022 to November 2024
Ran client and project delivery at a top-five global sustainability consultancy; owned QA and knowledge management for the Climate Resilience and Decarbonization business line.
- Led pursuit teams that won roughly $2M in engagements; closed two six-figure engagements sourced from own executive relationships.
- Led projects end to end from conception through delivery (client relationships, scope, budgets, CRM) for Fortune 500 and mid-market clients across retail, food and beverage, manufacturing, cloud services, healthcare, agriculture, and forestry.
- Designed Net Zero strategies integrating supplier engagement, product carbon footprinting, and life cycle assessment; built GHG inventories and Scope 3 mitigation assessments; implemented MRV systems aligned to ISO, Verra, Gold Standard, GHG Protocol, and SBTi.
- Delivered ESG reporting and disclosure support; facilitated executive stakeholder and supplier engagement.
- Managed and developed direct reports through mentorship, performance evaluation, and goal setting.
Director, Value Chain Certification · SustainCERT (Gold Standard)
Amsterdam, January 2020 to April 2022
Directed program design and certification at the creation of a new assurance category, owning policies, procedures, and decisions for verification of corporate Scope 3 insetting under the Gold Standard Value Change Initiative.
- Designed the certification methodology for corporate Scope 3 insetting, a category with no prior accounting rules, and wrote the policies and verification procedures still used in that market.
- Certified more than twenty land use and forestry projects, contributing to issuance of hundreds of thousands of credits with more than forty SDG co-benefits.
- Conducted GHG audit assessments under GHG Protocol guidance and ISO 14064.
- Advised ISEAL, Better Cotton, H&M, and Adidas on value chain and chain-of-custody design.
Earlier career, 1999–2020
Standards authorship (EcoLogo criteria for 300+ product categories, TerraChoice), nine years of ISO 14001, FSC, and USDA organic certification and audit practice, FSC audit leadership at Rainforest Alliance, and forest ecosystem simulation modeling as a research scientist at UBC and UQO, including climate scenario analysis for a million-dollar industry research collaboration. Details on request.
Technical
Languages and frameworks
PythonSQLSQLitePostgreSQLTypeScriptJavaScriptReactHono
Infrastructure
Cloudflare WorkersCloudflare PagesR2D1SupabaseGitHub Actions CI/CD
AI and data
Claude Code: multi-instance agentic buildFrontier LLMs: extraction & LLM-as-judgeLocal models: Llama/Ollama, Nomic, BERT/SBERTMulti-agent orchestration, resumable stateRAG & governed retrievalVision-transformer document ingestionChromaDBDuckDBSchema designETL from unstructured regulatory text
Standards
GHG Protocol: Corporate, Scope 3, Land Sector & RemovalsISO 14064ISO 14065ISO 17029ISO 14001ISO 14025SBTiCSRD & ESRSIFRS S2EU TaxonomyGRISASBCalifornia SB 54 & EPRGold StandardVerraFSCUSDA NOP