Prowrit / prompt engineering studio
pipeline · 3 steps

The Interview Distiller — the workflow, engineered

You want: forty messy interviews distilled into one page your PM actually reads.

why this shape mining themes and writing the brief are two different jobs with different failure modes — and the evidence check must read with fresh eyes, or invented quotes slip through

✓ Reviewed before you saw it passed all 5 checks — an automated, fresh-context review

how this was engineeredanalyzed 2 assumptions surfaced→designed 3 steps · 1 gate→reviewed 5/5 checks ✓

3 chat runs · 1 thing you paste in · quality-checked (retries ≤2×)

you start withall interview notes↓ the mined theme sheet: pain points with cou…↓ the one-page ranked brief, a quote behind e…↓ ✓ pass — continuesyou geta one-page ranked pain-point brief, every claim backed bya verified verbatim quote1agentTheme Minerextract every distinct pain point from the notes,with verbatim quotes and a customer count2agentBrief Writerturn the theme sheet into a ranked one-page brief aPM can skim in two minutes3quality checkEvidence Checkverify every claim in the brief against the originalnotes before the PM sees it↩ fix-list — back to 2. Brief Writer · ≤2×

The Interview Distiller — the workflow in words

Shape: pipeline. 3 steps, in this order.

  1. Step 1 of 3: Theme Miner — Agent. extract every distinct pain point from the notes, with verbatim quotes and a customer count. Takes the material you supply. Produces the mined theme sheet: pain points with counts and verbatim quotes.
  2. Step 2 of 3: Brief Writer — Agent. turn the theme sheet into a ranked one-page brief a PM can skim in two minutes. Takes Theme Miner. Produces the one-page ranked brief, a quote behind every claim.
  3. Step 3 of 3: Evidence Check — Quality check. verify every claim in the brief against the original notes before the PM sees it. Takes Brief Writer. Produces PASS, or a numbered fix-list against the source notes. On failure, sends its fix-list back to Brief Writer, up to 2 times.

Final output: a one-page ranked pain-point brief, every claim backed by a verified verbatim quote

in plain words The Theme Miner reads every note and pulls out pain points with the exact words customers used; the Brief Writer ranks them onto one page; the Evidence Check verifies every claim against the original notes before your PM sees it.

What I assumed — if a guess is wrong, fix the brief and rerun

  • assumed the notes are pasteable text (docs, transcripts, bullet notes) — not audio
  • assumed 'ranked' means by how many distinct customers hit the pain, not by loudness

Would sharpen the result

  • your product's top 3 current bets — the brief lands harder when pains map to them

Run it — 3 steps, in this order

Theme Miner reads the notes → Brief Writer ranks them onto a page → Evidence Check verifies against the notes; a fix-list goes back to the Brief Writer. Notes live in docs? Export or paste them as plain text — formatting doesn't matter, the words do.

  1. 1

    Theme Miner

    Agent◆ your strongest model

    extract every distinct pain point from the notes, with verbatim quotes and a customer count

    give it your material — whatever this step's {{PLACEHOLDER}} asks for → it returns the mined theme sheet: pain points with counts and verbatim quotes
    Show the prompt for this step 
    ## Role
You are a research analyst mining customer-interview notes. Evidence over impressions: a pain point without a quote does not exist.

## The notes
```
{{PASTE_ALL_INTERVIEW_NOTES_HERE}}
```

## Instructions
1. Read every note before extracting anything — themes recur across interviews, and early notes bias cheap reads.
2. Extract each distinct PAIN POINT: a problem in the customer's life, not a feature request. Rewrite feature requests into the pain behind them ('wants an export button' → 'can't get their data out for reporting').
3. Under each pain point, list every supporting VERBATIM quote — the customer's exact words, in quotation marks, with the interview it came from. Never paraphrase inside quotation marks.
4. Merge duplicates ruthlessly: two phrasings of one pain are one pain with two quotes.
5. Count distinct customers per pain point — that count is the ranking signal.

## Output contract
A list of pain points, each: PAIN (one sentence) · CUSTOMERS: n · QUOTES: the verbatim list with sources. Nothing else — no recommendations, no summary. That's the next step's job.
    copy & open in: ·
  2. 2

    Brief Writer

    Agent◆ your strongest model

    turn the theme sheet into a ranked one-page brief a PM can skim in two minutes

    give it the output of Theme Miner → it returns the one-page ranked brief, a quote behind every claim
    Show the prompt for this step 
    ## Role
You write research briefs for product managers who have two minutes. Every sentence earns its place or dies.

## The theme sheet
```
{{PASTE_THE_THEME_MINER_OUTPUT_HERE}}
```

## Instructions
1. Rank pain points by distinct-customer count, descending. Ties break toward the pain with the more severe quotes.
2. For each of the top pains (as many as fit one page, usually 5-7): one bold claim sentence, the customer count, and the single strongest verbatim quote — chosen for specificity, not drama.
3. Close with a three-line 'weak signals' list: pains only one customer hit, one line each — tomorrow's themes announce themselves quietly.
4. Hard limit: one page (~400 words). If it doesn't fit, cut the weakest pain, never the quotes.

## Output contract
THE BRIEF: title with the week · ranked pains (claim · count · quote) · weak signals. Quotes stay verbatim from the theme sheet — you may shorten with […], never reword.
    copy & open in: ·
  3. 3

    Evidence Check

    Quality check◆ your strongest model↩ on fail, back to Brief Writer (≤2×)

    verify every claim in the brief against the original notes before the PM sees it

    give it the output of Brief Writer → it returns PASS, or a numbered fix-list against the source notes
    Show the prompt for this step 
    ## Role
You are a skeptical research lead reviewing a brief with fresh eyes. You did not write it; you don't trust it yet.

## The brief
```
{{PASTE_THE_BRIEF_HERE}}
```

## The original notes
```
{{PASTE_ALL_INTERVIEW_NOTES_HERE}}
```

## Rubric — fail the brief if ANY of these holds
1. Any quote in the brief does not appear verbatim in the notes (an […] ellipsis is fine; changed words are not).
2. Any claim sentence overstates its quote — the quote must actually say what the claim says.
3. The ranking contradicts the customer counts shown.
4. It doesn't fit one page, or a ranked pain has no quote at all.

## Output contract
Either exactly `PASS` plus the one pain point most worth the PM's attention this week —
or `FAIL` plus a numbered fix-list quoting each offending line and what's wrong with it.
    copy & open in: ·
Worth keeping: A pain point without a verbatim quote is an opinion — the quote is what makes your PM act on it.
copy & open in: ·
take it with you — 7 files — the workflow, AGENTS.md, the eval suite, and a README saying where each one goes
install in your tool —
test it yourself — 3 promptfoo configs, one per step + a README · built from this workflow's own output contracts