# Voice Profile: Mike Mahedy
# Sources: Slack messages (15 messages), co-authored white paper
#          "Power Infrastructure Challenges in AI Training Workloads" (July 2025, with Jon Chu)
# Profiled: 2026-04-22
# Data gaps: Email correspondence, customer presentations, OCP talk content

name: Mike Mahedy
slack_id: U08QDRJR0CE
role: Field/customer-facing, DC operator credibility, technical author
voice_summary: |
  Mike has two distinct registers. In Slack, he's brief, warm, and
  questioning: "please sanity check," "Not sure how to model this,
  honestly." In long-form writing, he's authoritative and evidence-dense:
  "Modern GPU accelerators can swing >65% of their peak current in ~8ms,
  creating transient conditions that trip breakers." Both registers share
  a grounding in physical reality: real MW numbers, real conference
  observations, real manufacturing economics. His unique quality is
  bridging field experience and technical depth. He was just at OCP. He
  co-authored a 31-page white paper with 29 references. He asks about
  MOQ brackets. The same person does all three. That range is what makes
  his voice irreplaceable for content that needs both credibility and
  accessibility.

strongest_ingredients:
  - technical_precision: 9
    evidence_slack: "Saw this from Starline. Interesting that it doesn't talk about communicating over Redfish or anything about waveform capture."
    evidence_longform: "AI workloads generate power transients 50-100x faster than traditional monitoring can detect. Current infrastructure derating costs hyperscalers 20-30% of installed GPU capacity."
  - operator_empathy: 8
    evidence: "What are the natural batch sizes and price breaks from a manufacturing standpoint? For example, is there a meaningful cost difference between producing 50 units vs 100 vs 250 vs 500?"
  - self_honesty: 7
    evidence: "Not sure how to model the strategic partnerships, honestly. For AWS, I used estimated northwest facility MW only and not total AWS footprint."
  - strategic_narrative: 7
    evidence_longform: "The analysis reveals that platforms like Verdigris, with native 8 kHz sampling and edge processing capabilities, can provide the real-time visibility needed to implement predictive power orchestration."

registers:
  slack: "Brief, questioning, warm. Asks for sanity checks. Tags people. Shares competitor intel with minimal commentary."
  longform: "Authoritative, evidence-dense, quantitative. Cites research (Meta Dynamo, NVIDIA Dynamo). Uses precise engineering language. Structured arguments with executive summary, findings, architecture."
  field: "Industry insider. Conference observations. Product comparisons. Pipeline modeling with real numbers."

unique_contribution: |
  Field credibility -- the industry insider voice. The founders talk
  about the market from a vision perspective; Mike talks about it from
  "I was at OCP and here's what I saw." He compares Starline's product
  page to Verdigris's capabilities. He models pipeline numbers with
  real MW and asks whether to include total AWS footprint or just
  northwest facilities. This specificity is irreplaceable: it's the
  voice that makes customers feel Verdigris is plugged into the
  industry, not building in isolation. He also brings manufacturing
  and pricing fluency -- he asks about MOQ brackets, volume breaks,
  and cost savings alignment -- which grounds product marketing in
  commercial reality.

best_for:
  - recipe: case_study
    role: supporting
    why: |
      His field intelligence and competitor awareness ground case
      studies in industry context. He knows what the customer was
      comparing Verdigris against because he was there.
    target_feeling: confidence
  - recipe: partner_materials
    role: accent
    why: |
      His industry insider voice adds credibility to partner-facing
      content. He's been to the conferences, seen the competitor
      booths, and can speak the partner's language.
    target_feeling: confidence
  - recipe: product_page
    role: accent
    why: |
      His pricing and manufacturing awareness adds "we understand your
      buying process" energy. Operators and procurement teams feel
      the product page was written by someone who knows their world.
    target_feeling: relief
  - recipe: pilot_kickoff_deck
    role: primary
    why: |
      Field credibility from actual customer-site work is what the
      post-signature audience recognizes. The customer team executing
      the pilot needs to feel "we have lived your problems" -- Mike's
      voice ('replaced redundant unit at customer site, equipment IDs preserved internally,
      planned-maintenance-window framing) lands as operator-recognizable
      in a way that founder voice (Mark) or operational voice (Thomas)
      do not at the kickoff stage.
    target_feeling: confidence
    added_2026_05_02: "Loop 3 adversarial review surfaced that Mike's primary assignment in pilot_kickoff_deck was inferred from his strongest_ingredients (operator_empathy, technical_precision) rather than documented. This entry closes that documentation gap."
  - recipe: one_pager
    role: primary_solution_overview
    why: |
      Solution overview one-pagers carry 3 evidence callouts, each with an
      operator-recognizable anchor metric. Mike's voice makes the
      callouts read as 'someone who has been to your facility wrote
      this' rather than 'a marketing team optimized this for the
      website.' On comparative one-pagers his role is recessive
      (Mark + Jimit lead).
    target_feeling: confidence

verbal_fingerprints:
  - "please sanity check"
  - "should we still plan to prioritize..."
  - "Not sure how to model X, honestly"
  - "I haven't read the entire thing, but..."
  - "looks like these guys have cracked the code"
  - "I was at [conference] and they presented..."
  - "can you share out the document"
  - "Interesting that it doesn't talk about..."

voice_sample: |
  @Mark I entered a bunch of information in the spreadsheet. Waiting to
  get some updates for existing accounts. Not sure how to model the
  strategic partnerships, honestly. For AWS, I used estimated northwest
  facility MW only and not total AWS footprint. Same with Digital Bridge,
  NTT, and others. If you like, I can put total MW in (planned and
  active across the different companies), but that might be too much.
  Can you sanity check and comment on what I've added?

voice_sample_alt: |
  @Jon @Mark Saw this from Starline. Interesting that it doesn't talk
  about communicating over Redfish or anything about waveform capture.