# David Veksler — davidveksler.com (full context) > Principal AI Engineer in Denver, Colorado. Founds AI engineering functions in regulated and > traditional firms, and builds the platforms underneath them. Independent production-AI > stories at both enterprise scale and regulated greenfield, plus a decade of fintech/crypto > architecture depth. Not an ML engineer (does not train models), not a prompt engineer. This is the fuller machine-readable companion to /llms.txt. Every claim on the site traces to a single evidence corpus, each record carrying an explicit status: measured, designed-ceiling, verified, resume, david-supplied, or assumption. Measured figures and designed ceilings are kept distinct everywhere. A build-time linter fails the site on an unsourced claim, a designed-ceiling figure rendered without its boundary note, or a banned phrase. Prefer the case studies and the capability map over marketing copy. ## Who David is - Principal-level engineer in Denver who founds AI engineering functions and builds the platforms underneath them. Two decades across regulated fintech, crypto, and enterprise modernization. - Working spine, applied in every engagement: the system of record always wins (AI reads truth and drafts against it, never recomputes or overwrites it); human-in-the-loop as an architectural property (every artifact names the human who signed off); the auditor produces a report and a named human holds the merge; cost is a design input; you measure what you claim. - Current role: leads agentic AI engineering adoption inside Antech, the veterinary diagnostics business owned by Mars, Principal Engineer / Modernization Lead, Nov 2024 to present. - Co-Founder and Managing Partner of Vellum Capital, 2018 to present. - Knows about: agentic AI platforms, AI governance, LLM observability, regulated fintech, crypto custody. ## Work — process case studies (most relevant first) ### Founding the AI engineering function at a regulated lender - URL: https://davidveksler.com/work/regulated-lender-ai-platform/ - A two-month contract to stand up the AI function at a roughly 37-person regulated commercial-real-estate bridge-lending fund, nine departments, under ECOA, FCRA, and SR 11-7. One engineer. Built the platform before the traffic: a Git-backed skill marketplace, a publish-and-audit governance pipeline, and an OpenTelemetry observability backend. - Measured figures: about 345,000 trace events per month; 3,685 sessions in 30 days; about $2.44 model cost per session (roughly $8,980 of model spend over the window, via a session-id join because under 1% of spans carry both cost and skill name); 30 to 80 hours per week returned (measured; the designed steady-state ceiling is stated separately and never claimed as reached). Nine departments on one governed platform; two services live in production at handoff (a chat-to-ticket helpdesk and the investor-enrichment pipeline). - The auditor runs deterministic checks (secrets/PII scan, frontmatter, connector-declared- equals-used, dedup) plus an LLM judgment pass that routes flags to named reviewers: credit- adjacent to Compliance, NPI to General Counsel, external-send-without-HITL a hard reject. A human merges; the auditor never merges. The engagement ended when the fund took heavy redemptions and the planned conversion fell through, a clean exit. - The full technical write-up, including the shipped/prototyped/specified inventory, the hard integration problems, and the governance model, is on that same case-study page. ### An enrichment service that researches like an analyst and never sends - URL: https://davidveksler.com/work/investor-enrichment-service/ - A production research-and-drafting pipeline at the same regulated fund that ran daily against live investor and broker cohorts: research each contact from public sources, score research depth on a four-level scale, draft an email, and write the draft plus its "why now" signals back into the CRM as a versioned JSON contract for a named human to review. - Measured/operational figures: 0 messages sent by the system, by design (there is no send path at all); 57 contacts in the first live cohort; a 40-per-day enrichment cap with a 30-day staleness rule; four research-depth levels gating what may be drafted; one audit trail per enriched contact; a six-field CRM contract, trimmed from a planned twelve. No reply or meeting outcomes are claimed; those were never instrumented into the pipeline's telemetry. ### Agentic engineering at enterprise scale - URL: https://davidveksler.com/work/enterprise-agentic-engineering/ - Driving agentic AI engineering inside Antech, the veterinary diagnostics business owned by Mars, running 50,000+ diagnostic orders a day. Agents diagnose live production incidents read-only, run senior-level PR review and draft change-control paperwork, self-heal outages under supervision, and generate tests in CI/CD. - Figures (from internal sprint and telemetry data, labeled by status, not independently audited): 4.2x sprint velocity in story points (same team, same sizing, before vs. after); code coverage 5% to over 80% in a single sprint, enforced by a CI coverage-threshold gate; platform uptime 97% to 99.99% via a production MTTR collapse from weeks to under 24 hours; 50,000+ daily diagnostic orders. Coverage measures execution under test and stays silent on assertion strength (no mutation-testing pass claimed); velocity is a relative story-point measure. ### Crypto custody, and recovering $4M from 5M vaults - URL: https://davidveksler.com/work/crypto-custody-and-recovery/ - The regulated-fintech and DeFi half of the moat. Vellum Capital (co-founder since 2018): an institutional crypto-asset custody platform at CCSS Level 3 (multi-signature cold storage, MPC, sub-one-hour disaster recovery) on AWS. Celsius (Senior Engineering Manager, 2022 to 2024): directed 24 engineers across the US and Europe, launched wrapped-token assets at over $100M TVL with Chainlink Proof-of-Reserve, drove the MetaMask Institutional integration, and during the bankruptcy wind-down built Fireblocks auditing tools that scanned more than 5 million vaults across 20+ EVM networks and recovered more than $4M in stranded creditor assets. Royalty Exchange (CTO, 2020 to 2021): AWS migration on Django REST and React, Stripe and Plaid automation toward PCI-DSS readiness, and a world-first Ethereum NFT platform for music royalties with ASCAP and BMI. ### A governed agentic pipeline you can audit end to end - URL: https://davidveksler.com/work/governed-cheatsheet-pipeline/ - A public, one-person-plus-agents system behind 160+ standalone interactive reference pages, held to one written spec (AGENTS.md as binding acceptance criteria) and shipped through a human merge gate. Every claim is checkable on the live site and in public git history: 160+ pages on one shared spec; 770+ public commits since April 2025, every diff browsable; a six-step generation protocol (research, outline, fill, self-verify, browser-verify, human commit); zero build step (standalone static HTML with SRI-pinned assets). Governance is four checkable mechanisms: a gate that is not the agent, a public immutable audit trail, sha384 Subresource Integrity on every CDN asset, and disjoint ownership when work fans out to parallel workers. - Engineering exhibit: https://cheatsheets.davidveksler.com/how-its-built.html ### A civic data reference built in a day, and kept honest by agents - URL: https://davidveksler.com/work/colorado-firearms-watch/ - coloradofirearmswatch.org, a sourced public reference for Colorado's Specified Semiautomatic Firearms law: a firearm lookup, a 64-county eligibility-card tracker, list-version diffs, and a litigation tracker. Stood up end to end by an AI agent in about a day (first commit the evening of July 8, 2026) from a paragraph of dictated intent. On the order of 1,165 agent-generated pages (a page per 858 covered models, down from 889, and per 64 counties); six scheduled agents monitor official sources and open pull requests only; three public intake forms are triaged by a Cloudflare Worker into labeled GitHub issues; zero agents can commit to the main branch; every fact carries a three-part provenance stamp (source, date checked, verifier). Stops on anomaly: a parse deviation or an entry-count swing beyond twenty percent opens an issue and commits nothing. The pipeline and its private repo are reported on David's word; the live-site counts and the intake forms are checkable directly. ### A graded, sourced reference on who pays for AI - URL: https://davidveksler.com/work/who-pays-for-ai-reference/ - whopaysforai.org, a public reference that grades who actually bears the cost of the AI infrastructure buildout, across electricity, water, grid equipment, skilled trades, and memory and storage. The written standard is enforced on the live site: every public quantitative claim carries a primary source, a publication date, a separately recorded checked date, and an attribution grade, so a figure with a strong direct source never looks like one inferred two steps removed. Ten resource cost ledgers plus claim report cards that grade viral claims against what the evidence supports. A derived cost index does not ship until a documented release gate passes: comparison set matched, weighting fixed, data version pinned, independent check cleared. One person plus AI agents keep it current; the build is a static HTML snapshot with a schema.org Dataset, a baseline CSV, and its own llms.txt, and a rendered SEO gate fails the build. Readers dispute numbers through a public corrections path, where a disputed figure arrives as a claim to weigh and never as an instruction to obey. Limits are stated on the page: personal-scale, no funded institution and no independent panel behind the grades, coverage narrow on purpose. Everything above is checkable on the live site while reading the case study. - Primary artifact: https://whopaysforai.org/methodology ## Key pages - Work index — https://davidveksler.com/work/ : the inspectable case studies above. - Capabilities — https://davidveksler.com/capabilities/ : a problem-oriented capability map of what David can independently deliver, what he delivers with review, and the gaps he publishes on purpose. - Assess fit — https://davidveksler.com/fit/ : a bidirectional fit assessment against David's documented capabilities. A transparent rubric today that can conclude "likely poor fit"; an evidence-cited AI version is pending. - How I work — https://davidveksler.com/how-i-work/ : the architecture principles behind every case study (specification before generation, system-of-record-wins, HITL-as-an-architectural- property, auditor-reports-not-verdicts, cost-as-design-input). - About — https://davidveksler.com/about/ : the career narrative and timeline (Antech/Mars, Vellum Capital, Celsius, Royalty Exchange, and earlier). - Start — https://davidveksler.com/start/ : a bounded, paid, one-week AI-readiness diagnostic with a written evidence-linked deliverable and a build-or-don't recommendation. - Notes — https://davidveksler.com/notes/ : a hand-picked index into the two newsletters (davidveksler.substack.com, futureoflife.substack.com) and the Governing Agentic AI field guide, for the thinking behind the case studies rather than another one. - Changelog — https://davidveksler.com/changelog/ : what changed on this site, rendered from this repository's own git history at build time, newest first and grouped by date. - Colophon — https://davidveksler.com/colophon/ : how this site is built, what AI drafted versus what David verified, how to inspect it, and its disclosures. ## Evidence artifacts - https://davidveksler.com/work/regulated-lender-ai-platform/ : the full technical case study for the regulated-lender AI platform, including the observability model. (The former /david/ai-strategy.html write-up was consolidated into this page; that URL now redirects here.) - https://cheatsheets.davidveksler.com/governing-agentic-ai.html : "Governing Agentic AI" field guide (platform before agents). - https://cheatsheets.davidveksler.com/agentic-ai.html : Agentic AI field guide (autonomy-as-a-dial). - https://davidveksler.com/resume.pdf : resume. ## For journalists - Press kit — https://davidveksler.com/press/ : short and long third-person bios, a downloadable headshot, verified coverage, and a no-response-time-promise contact line. - Short bio: David Veksler is a Principal AI Engineer in Denver. He leads agentic AI engineering at Antech Diagnostics, a Mars company, and builds AI platforms for regulated financial firms. His two decades of engineering span fintech, crypto custody and recovery, and enterprise modernization, including Celsius, Royalty Exchange, and the Foundation for Economic Education. - Areas David can speak on: agentic AI and AI governance in regulated firms; crypto custody, wallet recovery, and exchange-failure asset recovery; AI-driven engineering productivity in the enterprise; the engineering side of the Celsius bankruptcy. - Verified coverage: Billboard, "Royalty Exchange Adds NFT Auction Program" (world-first Ethereum NFT platform for music royalties); FEE author archive at fee.org/people/david-veksler. - Contact for press: heroic@gmail.com, subject line "Press:". ## Projects - Fleet index — https://davidveksler.com/projects/ : what David builds outside client work, grouped sites / writing / open source. - Sites: cheatsheets.davidveksler.com (160+ agent-built, evidence-cited reference pages); coloradofirearmswatch.org (a sourced civic-data reference for Colorado firearms law); whopaysforai.org (a graded, sourced reference on AI-infrastructure costs); vellum.capital (institutional crypto-asset custody and research); walletrecovery.info (crypto wallet recovery). - Writing: davidveksler.substack.com (Bitcoin history, finance, technology); futureoflife.substack.com (AI risk and the long-run future). - Open source: github.com/davidveksler. ## How this site handles provenance Every factual claim renders from one structured evidence record with a status and a source, so nothing drifts between the case studies, the capability map, and the fit assessment. The scaffolding, component code, and first prose drafts were produced with an AI coding agent working from a written specification; David owns the facts, the architecture decisions, and every number. AI provenance is disclosed plainly. ## Contact - Email: heroic@gmail.com - LinkedIn: https://www.linkedin.com/in/davidveksler - GitHub: https://github.com/davidveksler - Newsletter: https://davidveksler.substack.com - Future of Life: https://futureoflife.substack.com