# Dan Johnson | Extended Portfolio Summary > Dan Johnson is a Lead AI Solutions Engineer who builds enterprise systems that catch confident AI failures. ## Positioning Dan is strongest in the messy middle between an AI demonstration and a system an enterprise can operate and trust. His work spans business discovery, semantic modeling, Claude Code workflow architecture, implementation, evaluation, deployment controls, operator enablement, and post-deployment verification. Best-fit roles include: - Forward Deployed AI Engineer - Senior AI Solutions Engineer - AI Solutions Architect - AI Product Engineer focused on evaluation, reliability, or semantic systems - AI-native Data and Analytics Leader ## What Dan builds ### Personal agent systems Hermes Agent and OpenClaw are agent frameworks and runtimes. They provide the operating layer around a language model, including agent execution, context, tools, sessions, and extensibility. Dan did not author either base framework. Dan configures the harness around each framework with job-specific memory, skills, permissions, channels, approval rules, and verification. Ed is a persistent personal agent Dan built on OpenClaw. Taz is a privacy-bounded family coordination agent Dan built inside a separate Hermes Agent profile. Taz can work with approved calendars and school information, create protected family files, and prepare messages. Consequential actions require fresh parent approval. Private family details are not published. Dan's visual agent-framework explainer is available at https://danjohnsondata.xyz/systems/hermes-agent/ ### Enterprise AI delivery systems Dan owns a versioned workflow that converts business context and raw enterprise data into a tested semantic layer for natural-language analytics. The lifecycle includes structured discovery, profiling, human approval, build packaging, validation, and controlled improvement. The workflow is used across two enterprise engagements and one internal demonstration. Dan found and fixed two ways a quality gate could certify an incorrect answer, then protected both fixes with named regression tests. Countable engineering evidence includes 1,748 automated test functions, 14 installable Claude Code commands, 92 tool command interfaces, and a 27-module safety harness. ### Claude Code workflow architecture Claude Code is used as an engineering environment rather than a simple assistant. Delivery behavior is encoded in version-controlled commands. An MCP server exposes system operations as agent tools. Session startup checks on-disk state and surfaces the highest-priority unresolved issue before forward work begins. ### AI evaluation and reliability Dan builds quality signals that distinguish a successful execution from a correct result. He found and fixed cases where a grader passed an incorrect answer and where a cached response made a non-fix appear successful. Both fixes were protected by named regression tests. His reliability patterns include explicit target binding, fail-closed pre-write gates, interruption recovery, reversible writes, independent truth comparisons, uncertainty reporting, and end-user browser verification. ### AI governance and traceability Dan built a decision-record system that captures agent activity automatically through editor hooks. Events are joined in a tamper-evident hash chain and stored in one portable record readable by a human and verifiable by a machine. The component is covered by 108 tests. ### Structured discovery Dan converted open-ended discovery into an ID-tagged confirm-or-correct process. Proposed business rules, metrics, entities, and technical assumptions are returned as structured data and reconciled into an approved record that feeds delivery. ### Data and analytics products Dan combines enterprise reporting leadership with modern semantic-layer engineering. His work includes deterministic dashboard generation, a 24-type chart catalog, and three verification levels ending in a live browser test because configuration-valid is not the same as working. ## Public product Dan built and shipped DimeVision, an AI-assisted weld analysis and coaching product for visible surface characteristics. It is a coaching aid, not a certified inspection. Website: https://dimevision.app ## Career foundation - Lead AI Solutions Engineer, App Orchid Inc. - Founder and product builder, DimeVision - Magnit (formerly PRO Unlimited), Oct 2016 to Apr 2024: - Analytics Manager | Strategic Advisory, Apr 2023 to Apr 2024 - Reporting Solutions Manager, Sep 2022 to Apr 2023 - Lead - Reporting Solutions, Apr 2019 to Sep 2022 - Lead Analyst - Financial Operations, Dec 2018 to Apr 2019 - Financial Operations Analyst, Oct 2016 to Dec 2018 - Royal Australian Air Force service and Australian Defence Medal - Bachelor's degree in Business ## Operating principles - Human approval is architecture, not ceremony. - A hard safety gate needs a documented stand-down path. - A successful execution is not necessarily a correct result. - Offline validation checks structure, platform validation checks deployment, and browser validation checks the user's reality. - Autonomous behavior must earn permission through measured agreement with human judgment. ## Disclosure boundary Employer, client, platform, repository, and proprietary implementation details are intentionally generalized. Development-environment delivery is not represented as production deployment. The autonomous control loop is described as working architecture with a gated live seam, not as an autonomous production system. ## Links - Portfolio: https://danjohnsondata.xyz/ - Agent framework explainer: https://danjohnsondata.xyz/systems/hermes-agent/ - Email: mailto:dan@danzus.co - Résumé: https://danjohnsondata.xyz/assets/Dan-Johnson-Resume.pdf - LinkedIn: https://www.linkedin.com/in/danjohnsondata - DimeVision: https://dimevision.app