// Variant: main — apex mathur.dev. Copy mirrors Resume/main.tex (assets/main.pdf).
// Role subdomains (ai/aipm/fde/sde) keep their own files and are not affected.
(window.RESUME_CONTENT = window.RESUME_CONTENT || {}).main = {
  meta: {
    role: 'main',
    docTitle: 'Arpit Dev Mathur — Software & AI Engineer',
    resumePdf: 'assets/main.pdf'
  },

  hero: {
    eyebrow: 'Software & AI Engineer · San Francisco',
    tagline: (
      <>
        Software engineer with 6+ years building production systems at Windows scale — now
        applying that same rigor to production <em>RAG, evaluation, and multimodal AI</em> systems.
      </>
    )
  },

  // About has no counterpart in main.tex; same copy as the ai variant.
  about: {
    lead: "I'm a software engineer now building AI systems.",
    paragraphs: [
      <>I spent six years at Microsoft on Windows core platform problems (static analysis,
        cross&#8209;platform .NET, and a rollback framework used during live OS regressions), then
        went back to school to go deeper on ML.</>,
      <>These days I'm most interested in the seam between classical software engineering and
        modern ML: evaluation harnesses, retrieval pipelines, and the operational glue that
        makes LLM&#8209;based products reliable enough to ship.</>,
      <>Off&#8209;screen: hikes, weightlifting, growing plants.</>
    ]
  },

  roles: [
    {
      when: { start: 'Oct 2025', end: 'Jul 2026' },
      title: 'AI Engineer',
      company: 'Asurion',
      loc: 'San Francisco, CA',
      projects: [
        {
          bullets: [
            <>Engineered a modular, configuration&#8209;driven <b>RAG pipeline</b> using Redis hybrid search, Gemini embeddings, and ZeRank reranking, improving retrieval precision by <b>14.8%</b> and enabling no&#8209;code experimentation.</>,
            <>Automated <b>RAG evaluation</b> across customer&#8209;specific golden datasets with custom&#8209;built metrics and benchmarked BGE, Qwen3, Nemotron, and Gemini embeddings to drive production model selection.</>,
            <>Built an async, multi&#8209;provider <b>vision service</b> exposed as an MCP tool to the production chatbot's agentic LLM harness, classifying customer uploads by request type (damage, screenshot, error code etc.) at <b>95.5% accuracy</b> and cutting image payload size by <b>37%</b>, with OCR and NSFW safeguards.</>
          ]
        }
      ]
    },
    {
      when: { start: 'Aug 2016', end: 'Jan 2023' },
      title: 'Software Engineer',
      company: 'Microsoft',
      loc: 'Redmond, WA',
      projects: [
        {
          name: 'Retrodict',
          kicker: 'Static code analysis',
          bullets: [
            <>Developed a <b>static code analyzer</b> that identifies unsafe code patterns in the Windows OS codebase.</>,
            <>Analyzed <b>100s of Windows binaries</b> and presented results in an easily triage&#8209;able format using SARIF (Static Analysis Results Interchange Format).</>,
            <>Saved over <b>$1 million</b> in potential bug bounty payouts and development costs by identifying and resolving <b>100+ security issues</b> using the tool.</>
          ]
        },
        {
          name: 'Windows Presentation Foundation',
          kicker: 'WPF',
          bullets: [
            <>Migrated WPF to <b>.NET Core</b>, porting <b>over a million lines of code</b> and enabled WPF apps to be OS&#8209;agnostic, saving app developers time on every OS upgrade.</>
          ]
        },
        {
          name: 'Known Issue Rollback',
          kicker: 'KIR',
          bullets: [
            <>Developed a <b>Velocity&#8209;based known&#8209;issue rollback (KIR) framework</b> to enable selective rollbacks (without an update cycle) if regressions are discovered in Windows OS code.</>,
            <>Built an <b>Azure&#8209;based system</b> to remotely modify group policy and selectively roll back Windows OS changes, reducing customer impact by <b>over a month</b> during critical regressions.</>
          ]
        }
      ]
    },
    {
      when: { start: 'May 2015', end: 'Aug 2015' },
      title: 'Software Engineer Intern',
      company: 'Microsoft',
      loc: 'Redmond, WA',
      projects: [
        {
          bullets: [
            <>Prototyped and developed a <b>voice&#8209;activated send&#8209;email</b> feature in Cortana, allowing end&#8209;to&#8209;end hands&#8209;free email functionality.</>,
            <>Added feature to proactively launch Cortana to dictate certain emails (when received from a list of selected senders) and allow the user to choose to reply or leave the email as unread.</>
          ]
        }
      ]
    }
  ],

  education: [
    { when: 'May 2026', school: 'University of San Francisco', degree: 'M.S., Data Science & Artificial Intelligence' },
    { when: 'May 2016', school: 'University of Illinois at Urbana‑Champaign', degree: 'B.S., Computer Science' }
  ],

  // Group order follows main.tex.
  skills: [
    {
      cat: 'Languages',
      items: [
        { name: 'Python', primary: true },
        { name: 'SQL', primary: true },
        { name: 'C#' },
        { name: 'C++' }
      ]
    },
    {
      cat: 'LLM Systems',
      items: [
        { name: 'RAG', primary: true },
        { name: 'Embeddings', primary: true },
        { name: 'Vector Search' },
        { name: 'Re-Ranking' },
        { name: 'MCP' },
        { name: 'Tool-Calling' },
        { name: 'Orchestration' },
        { name: 'Structured Outputs' },
        { name: 'Evaluation Pipelines', primary: true }
      ]
    },
    {
      cat: 'ML',
      items: [
        { name: 'PyTorch' },
        { name: 'Scikit-Learn' },
        { name: 'NumPy' },
        { name: 'Pandas' },
        { name: 'Deep Learning' },
        { name: 'Classification' }
      ]
    },
    {
      cat: 'Cloud & Infra',
      items: [
        { name: 'Azure', primary: true },
        { name: 'AWS' },
        { name: 'GCP' },
        { name: 'Docker' },
        { name: 'FastAPI' }
      ]
    },
    {
      cat: 'Data Engineering',
      items: [
        { name: 'PySpark' },
        { name: 'ETL' },
        { name: 'Spark' },
        { name: 'MongoDB' }
      ]
    }
  ]
};