Lead Data Architect

The user wants me to reformat this job posting into clean semantic HTML. Let me analyze the content and structure it according to the rules. First, I need to identify the sections: 1. About the role / Company description (the first paragraph about Domino's) 2. Job Description (the "As Lead AI Architect..." paragraph) 3. Responsibilities (multiple sub-sections) 4. Qualifications 5. Preferred Qualifications 6. Benefits (mentioned at the end) The rules say: - Use

for section headings - Use

• for lists - Use

  for prose - Don't repeat job title as heading - Don't put Job ID/requisition numbers as headings - Preserve ALL substantive information - Remove boilerplate (cookie/navigation/legal/EEO/scam-warning) - Include section ONLY when real content exists - No filler/placeholder text - Output ONLY HTML fragment Looking at the content: - The first paragraph is company description/about the role - "Job Description" paragraph is the role overview - "Responsibilities" has multiple sub-sections with detailed bullet points - "Qualifications" has a list of requirements - "Preferred Qualifications" has a list - "Additional Information" mentions Benefits but only partially listed (cut off) I need to structure this properly. The responsibilities section has multiple sub-sections: Architecture Ownership and Direction, Semantic Search, Retrieval and RAG Architecture, Generative AI Portfolio, Agentic AI Portfolio, Governance, Trust and Security, Observability, Economics and Platform Engineering, Technical Leadership. These could be sub-headings under Responsibilities, but the rules say use

  for section headings. I could make each a separate

  or keep them as sub-sections under Responsibilities. Since they are distinct areas, I'll make them separate

  sections. But the rules say "use

  for section headings (e.g. About the role, Responsibilities, Requirements, Qualifications, Skills, Benefits, Pay, Schedule)". So I should have main sections: About the role, Responsibilities, Qualifications, Preferred Qualifications, Benefits. However, the responsibilities are very detailed with sub-sections. I could either: 1. Make each sub-section a separate

  (e.g., "Architecture Ownership and Direction", "Semantic Search, Retrieval and RAG Architecture", etc.) 2. Keep them as sub-headings within Responsibilities using

  or just bold text, but the rules only mention

  for section headings. The rules say "use

  for section headings" and give examples. It doesn't forbid using

  for sub-sections. But the examples are top-level sections. I think it's better to keep the main sections as

  and use

  for sub-sections, but the rules don't mention

  . However, the instruction says "use

  for section headings" - it doesn't say exclusively. But to be safe, I'll make each major responsibility area a separate

  section, as they are substantial. But the user said "Prefer a few substantive sections over many thin ones." So maybe keep Responsibilities as one section with sub-sections as

  or just structured lists. However, the content is very detailed and each sub-section has multiple bullet points. I think it's appropriate to have them as separate sections. Let me check the content again: The responsibilities are grouped under headings like "Architecture Ownership and Direction", "Semantic Search, Retrieval and RAG Architecture", etc. Each has multiple paragraphs and bullet points. They are substantial. I'll structure as: -

  About the role

  with company description and job description paragraph -

  Responsibilities

  with sub-sections as

  ? But rules only mention

  . I could use

  for each responsibility area. That would be many sections but they are substantive. Alternatively, I could have one Responsibilities section with the sub-sections as

  but the output is HTML fragment, so using

  is fine. The rules don't prohibit other heading levels. They just say "use

  for section headings". I'll use

  for main sections and

  for sub-sections within Responsibilities. But the instruction says "Output ONLY the HTML fragment — no /, no markdown code fences." So I can use any valid HTML. I'll do: -

  About the role

  -

  Company description...

  -

  Job description paragraph...

  -

  Responsibilities

  - Then for each responsibility area:

  Area Name

  followed by
  • or

    as appropriate. But the rules say "Use

    • for any list of duties/requirements/benefits, and

      for prose." The responsibilities are written as prose paragraphs, not bullet lists. However, they are detailed descriptions. I could keep them as

      or break into bullet points. The original uses paragraphs. I'll keep as

      for prose, but some are lists (e.g., "Chunking, embedding and indexing: design chunking strategy..."). That's a list of duties. I'll convert to bullet points where appropriate. Actually, looking at the responsibilities, each sub-section has multiple sentences describing duties. They are not formatted as bullet points in the source. But the rules say "Use

      • for any list of duties/requirements/benefits". So I should convert duties to bullet points. However, the source is prose. I need to parse each duty. This is tricky. The responsibilities are written as paragraphs with multiple responsibilities in each. For example, "Architecture Ownership and Direction" paragraph has multiple sentences each describing a responsibility. I could split into bullet points. But the instruction: "Preserve ALL substantive information; do NOT invent or summarize away detail." So I must keep all details. I'll keep the prose as

        for each paragraph, but if there are clear lists, use

        . Looking at the text: "Own the enterprise AI reference architecture end to end, establishing scalable standards, principles, and design patterns across business groups." That's one duty. "Own the roadmap for agentic systems platforms..." another. So I can split by sentences. But some sentences are long with multiple clauses. I'll keep as paragraphs for readability, but the rules say use
        • for lists of duties. The example headings include "Responsibilities" as a section heading, implying a list of duties. I think the best is to treat each sub-section as a subsection with its own heading (h4) and then use bullet points for the duties within. However, the source doesn't have bullet points; it's paragraphs. I'll convert each distinct responsibility into a bullet point. Let me parse each sub-section: 1. Architecture Ownership and Direction: - Own the enterprise AI reference architecture end to end, establishing scalable standards, principles, and design patterns across business groups. - Own the roadmap for agentic systems platforms, translating business objectives into sequenced technical strategy, milestones, and a build-depth decision for every use case with explicit, traceable reasoning. - Design and maintain the paved road for AI development – golden patterns for building, serving, evaluating, and integrating models and agents, with an API-first integration default, MCP by exception against explicit criteria, and guardrails inherited rather than bolted on per team – so teams ship without re-solving solved problems. - Guide optimization of AI compute and token usage as adoption scales, and lead build-versus-buy, model, and framework decisions with evidence – benchmarks, cost per answer, real-load latency. 2. Semantic Search, Retrieval and RAG Architecture: - Chunking, embedding and indexing: design chunking strategy per content type and tune size, overlap, and boundaries against measured retrieval quality; select and evaluate embedding models; own re-embedding and versioning. - Query construction and hybrid retrieval: implement query rewriting, decomposition, and expansion; combine dense and sparse retrieval with fusion and re-ranking; manage the recall-versus-precision and latency trade-off explicitly. - Retrieval and response evaluation: instrument recall, precision and faithfulness against curated golden datasets; ground answers in the certified semantic layer and knowledge graphs where they outperform vector retrieval; design for graceful abstention over confident wrong answers. 3. Generative AI Portfolio: - Define prompt and instruction management standards – versioning, templating, context budgeting, and regression testing on prompt or model changes. - Own the AI gateway and multi-model routing, matching each request to the right model on complexity, cost, latency, and quality, and own the benchmarking discipline for how models are compared, selected, and re-tested as the landscape shifts. - Establish the evaluation harness as a launch gate – offline/online evaluation, LLM-as-judge with human calibration, A/B and shadow testing, and an explicit quality threshold we won't launch below. - Hold a clear, evidence-based position on when prompt engineering suffices versus when fine-tuning or customization is warranted, and define caching strategy as a deliberate cost and latency lever. 4. Agentic AI Portfolio: - Architect the orchestration layer that chains agents, tools, retrieval, and human review – orchestration as the enforced contract, domain agents scoped by journey and business context, and a reusable skills and tool layer where reusability actually lives. - Design memory and context management with an explicit position on what persists, for how long, and under what access rules; define enterprise standards for tool/function calling and MCP server contracts. - Establish agent guardrails – authorization scope, permission models, approval and escalation paths, sandboxing, fallback behavior, and safe failure – plus step-level tracing, regression testing, audit, and incident response. - Define the agent registry, tool catalog, memory and state services, and lifecycle from prototype to monitored production asset, and apply agentic patterns – spec-driven development, AI-assisted code and test generation – to our own engineering practice. 5. Governance, Trust and Security: - Define responsible AI standards – bias/fairness testing, transparency, explainability, and human oversight on high-risk outputs – aligned with enterprise data governance policy rather than parallel to it. - Establish AI security patterns for prompt injection, data exfiltration, jailbreaks, and model/supply-chain risk; partner with the Data Governance team on classification, stewardship, certified sources, and access controls – Microsoft Purview, Entra ID, Unity Catalog, row-level security, PII handling, and retention. - Classify use cases by risk, drive attestation with Security, Legal, Privacy, and Risk, and lead third-party model/vendor technical review, translating findings into architecture constraints. 6. Observability, Economics and Platform Engineering: - Instrument tracing, telemetry, and drift signals from day one so agent behavior is observable before it's questioned; build cost attribution and drive cost per successful answer as a first-class metric. - Contribute to the instrumentation that ties AI investment to measurable outcomes – partnering with platform and finance teams so unit economics are a first-class design constraint. - Define the Azure-native platform stack and lifecycle with MLOps and Engineering – Azure AI Foundry, Azure ML, AKS, model/agent registry, CI/CD, landing-zone and environment strategy, Entra ID and network patterns, capacity, and LLMOps/AgentOps for versioning, promotion, and rollback. - Partner with Engineering and Architecture teams on data readiness, certified metrics, lineage, and data contracts, and with Engineering on turning reference patterns into shipped services – defining the sanctioned builder environment so teams move fast without bypassing controls. 7. Technical Leadership: - Raise the bar across Engineering and solution architecture through design reviews, system-level reasoning, and mentorship – driving rigor without direct reporting relationships. - Translate architecture trade-offs into clear, decision-oriented narratives for senior technology leaders and peer architects, and lead cross-team decisions with partners such as Microsoft, Databricks, strategic AI vendors, and systems integrators. - Write architecture decision records,