Skip to content
Brian Kudera

Generalized · employer-safe

Brian Kudera

Data Engineering · Business Intelligence · Data Architecture

Download TXT

A deliberately generalized overview. I keep specific employers, customers, and internal systems off the public web — what follows is the shape of the work and the principles behind it. For specifics relevant to a particular role, get in touch.

Summary#

Senior AI and data engineer with 20+ years of experience designing enterprise data platforms, governed analytics systems, and production automation. Combines deep expertise in Snowflake, dbt, SQL, Python, data architecture, and multi-tenant SaaS systems with hands-on experience developing AI-assisted engineering workflows and agentic orchestration.

Experienced building systems around large language models using structured workflows, reusable tools, human review gates, testing, observability, and software engineering discipline. Translates complex operational requirements into scalable data and AI solutions that users can understand and trust.

AI and agentic systems#

  • LLM application architecture · agentic workflow design · agent orchestration
  • AI-assisted software development · tool and function calling · structured outputs
  • Prompt and context engineering · human-in-the-loop approval workflows
  • AI evaluation and validation · guardrails and failure handling
  • LLM observability and tracing · OpenTelemetry · Claude Agent SDK

Data and platform engineering#

  • Python · SQL and T-SQL · Snowflake · dbt Core
  • Data modeling and semantic architecture · multi-tenant data platforms
  • Data quality and validation · incremental data processing
  • APIs and integration services · GitHub and CI/CD

Application and cloud development#

  • TypeScript · Next.js · Azure · Docker
  • PostgreSQL · REST APIs · GitHub Actions

Selected AI project#

Agentic Data Engineering Toolkit#

Architected an AI-assisted software delivery framework for complex analytics-engineering projects. The toolkit combines specification-driven agent workflows, 60 versioned reusable skills, repository-aware tools, automated validation, retrospective analysis, and OpenTelemetry tracing. It is designed to augment engineers while preserving explicit approval, testing, and architectural controls.

  • Developed a wrapper around the Claude Agent SDK to standardize agent execution, tool access, workflow instructions, error handling, and telemetry across engineering use cases.
  • Built AI-assisted workflows for requirements grooming, implementation, technical retrospectives, and code review, with reusable declarative workflow definitions in Markdown and structured configuration.
  • Established human-in-the-loop checkpoints, validation gates, and fallback patterns for nondeterministic AI workflows; kept deterministic automation in charge where it was safer.
  • Integrated more than 16,000 existing and purpose-built validation checks into agent-driven preflight and delivery workflows.
  • Instrumented 565 agent sessions and 26,478 tool invocations, achieving a 97.2% tool-execution success rate; 96.4% of input tokens were served from cache in the measured environment.
  • Supported 95 Jira tickets through the fully instrumented agent workflow across 11 contributors, 6 product lines, and 3 repositories.
  • Contributed to a 3.0× increase in merged PR throughput at unchanged headcount following adoption of the agent toolkit. Average review threads per PR fell 79% from the May 2026 peak to July 2026 while monthly PR volume increased from 38 to 45.

Selected data and platform work#

  • Modernized a legacy BI environment in a multi-tenant SaaS analytics context — consolidating duplicated, conflicting metrics into a single canonical layer so teams stopped arguing about whose number was right.
  • Architected governed, historically accurate data models using conformed dimensions and type-2 change tracking, creating trusted enterprise data foundations for analytics, forecasting, and AI applications.
  • Developed automated data-quality and business-rule validation frameworks that protect downstream analytics and AI systems from incomplete, inconsistent, or unauthorized data.
  • Designed the data and trust model for a production platform serving a community nonprofit, including a shared inbox on the organization's own domain and a self-organizing photo library. (See The Nonprofit Back-Office.)
  • Built and deployed custom, fast, SEO-friendly websites for small organizations, delivered with an owner's manual so non-developers can maintain them. (See Custom Website Building.)

AI engineering judgment#

  • Defined architectural boundaries between deterministic automation and LLM-driven reasoning.
  • Established review, validation, observability, and fallback patterns for nondeterministic workflows.
  • Evaluated build-versus-adopt decisions for emerging agent frameworks and orchestration technologies.
  • Translated agent execution failures into durable improvements to workflow design, tools, instructions, and engineering standards.
  • Partnered with engineering and product leadership to identify high-value, operationally appropriate AI use cases.

Engineering principles#

  • Trust is the product of accuracy, explainability, and security — zero on any axis zeroes the whole.
  • A metric with no decision attached is not a metric.
  • Fail loudly and locally; never quietly and globally.
  • Automation without explanation is just a faster way to be wrong at scale.
  • Human accountability is part of the architecture, especially when a workflow is nondeterministic.
  • AI should be constrained by explicit tools, validation, observability, and approval boundaries.
  • The customer is always somewhere in the data model.