Building product sense at HelloPM · Assignment 2 · Cohort 51 · Solo PM and builder, 0 to 1
Kural
Draft emails in seconds that still sound like you
The live prototype: type a rough note and let Kural draft it. Open in a new tab ↗
Problem statement
Working professionals who already use AI email tools struggle to send high-stakes emails to customers and managers without rewriting the drafts, because the tools polish sentences but do not read people: one voice, one register, every time. 71% of AI users surveyed cannot send AI drafts as written, and 22% have quietly abandoned the tools.
Kural reads the situation first, urgency, power dynamic and emotional register, then drafts in your own voice.
Highlights
- 71%
- Cannot send AI drafts as is
- 42%
- Say tone is the hardest part
- 36 + 4
- Survey responses and interviews
- v1.2
- Live prototype
Why Polished Drafts Still Get Rewritten
The P0 problem
Professionals rewrite the drafts they are given
Even people who already use AI email tools rewrite what they get: the tools polish sentences but do not read people.
of AI users surveyed cannot send AI drafts as they are
have quietly abandoned the tools
The root cause. Generic assistants hold no persistent model of your voice and no structured read of the situation.
View project brief, origin & roleHow it started, the assignment and my role
Origin
How it started
The brief. Stop deconstructing products and start building one: pick a product I use often, find one real friction, and prototype an AI fix.
The friction. An AI assistant hands me a clean draft and I rewrite it anyway. The sentences are correct but not mine, and they read the same to a customer as to my manager.
The gap. Between a polished draft and a sendable one. That gap became the project.
What I built. Kural classifies who you are writing to and why, holds a persistent model of your voice, and drafts something you can send without a rewrite. I built it solo and shipped it at kural-iota.vercel.app.
The assignment
Design and prototype an AI-powered feature
The ask. Act as the PM of a product I use often and find one meaningful opportunity to improve it with AI.
My scope. One P0 problem and a demonstrable prototype.
The requirement. Validate the thinking with PMs from that industry, outside the assignment.
Role and build
Solo, from spec to shipped
Solo product manager and builder, end to end
Live prototype, v1.2
36-person survey and four interviews
How it was built. I wrote the spec and directed Claude Code to generate the Next.js app end to end, then tested it myself before shipping it on Vercel.
Proving Voice Is the Blocker, Not Grammar
Survey and interviews
Where the time and friction go
I ran a 36-person survey through friends, WhatsApp and Reddit, plus four interviews, asking where the time and friction in an important email actually go.
Voice is the blocker, not grammar
of AI tool users cannot copy-paste the draft. It sounds like a robot and lacks their voice.
Tone is situational
say tone is the hardest part of an important email. A customer, a manager and a professor each need a different register.
The work hides in wording
spend five minutes or more on one important email. A third lose it to wording and tone, a quarter to second-guessing.
Validation
Cross-checked against published research
A 36-person survey finds a signal; it does not prove one, so I held each finding against larger published studies. The survey and the literature agreed.
of the professional workweek goes to email
of readers distrust copy that reads as bot-written
accept it once it feels human
My own survey matched. The voice problem is a trust problem, and the friction is specific and measurable.
| Finding | My survey | Published |
|---|---|---|
| Email is a heavy time cost | ✓ | ✓ |
| Bot-sounding copy is rejected | ✓ | ✓ |
| Tone is the hardest part | ✓ | ✓ |
| Human-sounding copy is trusted | ✓ | ✓ |
Meet Kural
The product
Rough note to sendable draft
Your thoughts
Subject: Sick Leave Request for Today
Hi Priya,
I'm not feeling well today and would like to request a sick day to rest and recover. I'll check in on anything urgent and plan to be back tomorrow.
Thanks for understanding,
George
How Kural Reads the Situation Before It Writes
The engine
Classify first, compose second
Tone is a decision, not an accident. The schema settles the situation before a single line is drafted.
-
01
Classify
Urgency, power and emotion are set before any draft.
-
02
Adapt
Upward, customer or peer, each reader pulls its own register.
-
03
Draft
An anti-cliche guard strips stock AI openers before they ship.
-
04
Learn
Style memory learns from your changes, so later drafts sound more like you.
What Existing Tools Leave Open
Competitors
The gap Kural takes
Existing tools solve speed, cost, or integrations—leaving the voice-and-situation gap completely untouched.
| Feature | SuperhumanAsks you to replace your whole inbox | EllieVoice-trained tier is the expensive one | Gemini in GmailStays generic unless you prompt it hard every time | KuralCovers all four checks below |
|---|---|---|---|---|
| Strength | Fast | Can train on your voice | Cheap and close at hand | Voice and situation built in |
| Voice-first by default | ✗ | ✗ | ✗ | ✓ |
| Reads the situation | ✗ | ✗ | ✗ | ✓ |
| No inbox switch needed | ✗ | ✓ | ✓ | ✓ |
| Voice without a premium tier | ✗ | ✗ | ✗ | ✓ |
How I Measure Success
Metrics
One number matters: the send-as-is rate
A draft counts as a win only when the sender ships it with barely a touch.
of drafts sent as is, edit distance under 10%
regenerations per draft, ask-gate under 25%
P95 draft latency
Next project
Kittii: led discovery through MVP development at an AI lost-and-found startup with a team of eight, then planned a phased go-to-market.