🔒 Fidelity check is locked

This audio↔story breakdown is insider-only. If you know the key, add it to the URL — if not, ask around. 😉

← back to the story

🎧 Audio ↔ 📄 Story · Fidelity Check

How close is the page to Venky's voice?

Comparing the 16:37 recording (auto-transcribed, ~2,600 words) against the published The Prologue (~1,120 words).  ·  ← back to the story

0/100faithfulness
Faithful retelling.

Every story beat Venky tells in the recording survives on the page — Pondicherry, the two-day Voice IQ build, "finding it a blessing," the passport number, the AWS Conclave. The page is roughly half the spoken length because it tightens his long, looping delivery into clean prose, but nothing is invented and nothing material is dropped.

Where the score comes from

Story beats preservedDid every moment Venky told make it onto the page?90%
Facts & quotes accuracyNumbers, names, dates, and direct quotes88%
Tone & voice matchDoes it still sound like Venky telling it?88%
Verbatim wordingWord-for-word overlap — low by design for a written piece13%
Length retainedStory is 43% the length of the spoken version43%

▮ green = meaning preserved · ▮ amber = minor drift · ▮ grey = expected difference (rewrite, not loss)

✓ Kept faithfully

  • Pondicherry, Feb 2025 — Sri had moved there; the whole team off-site, two-to-three days
  • Art of the possible — mini-hackathons, diverge/converge, dark mode, ship something small
  • Change management in the tech team — the insecurity, two tracks, the safe-space tension
  • Voice IQ's genesis — Venky + Sreenath, Cursor & Lovable, made a logo, named it
  • Sri's brief — "don't overthink it… just give them the transcript" (near-verbatim)
  • The manual Gemini workflow & "we went deeper" / the kida itch
  • Architecture — Talkdesk, Gemini on Vertex AI, coaching prompts, dashboard; productionised under Sreenath
  • "Finding it a blessing" — the 7th-May group DM demo (verbatim phrase)
  • Data pipeline / BigQuery — "garbage in… wisdom in, wisdom out," single source of truth
  • Passport number — the PII scare, InfoSec early; AWS Conclave (Voice IQ + TC Copilot)

~ Changed / dropped

  • "PN" productionised it → page names Vivek "The Builder" (recording only says "PN")
  • Ostara / Europe Express proposal-generation example — dropped from the page
  • Yogi's coaching-prompt help — recording credits him on the prompts; page only notes he was "left in the dark"
  • Bureaucracy / "not allocated to ET" / "too many cooks" aside — compressed to "no budget, no mandate"
  • "ED team" / "Artemis team" framing — generalised to "the whole team"
  • Repetition & tangents — the looping, self-correcting spoken asides are smoothed out
  • Tightened, not lost: the 45-min-call → scored-summary point lands more crisply on the page than in the recording
Read it like this: the big number is faithfulness of meaning, not transcription accuracy. A score near 90 means "trust this page as Venky's story." The 13% verbatim and 43% length bars are healthy — Venky speaks in long, looping sentences, and the page distils that into half the words without losing a single beat. If verbatim were near 100, the page would just be a transcript, not a voice.

Method. Audio transcribed locally with whisper.cpp (base.en). Verbatim (13%) and length (43%) are measured directly (difflib sequence ratio & word counts). Beats, facts, and tone are a beat-by-beat human/AI reading of transcript vs. page. Overall faithfulness is a weighted blend (beats 40% · facts 35% · tone 25%); verbatim is reported separately, not folded in.