Dilith Dinesh
Three clicks. You're either very bored or very thorough. Either way, hello.
Product operator at the boundary between customer reality and product systems.
I translate what breaks in the operational layer into capabilities that scale.
- NOW
- Associate Product Manager, VoiceStack. Voice AI for dental practices.
- BEFORE
- Two years at Zuper. Built the roofing vertical from zero.
VoiceStack · Healthcare Voice AI · Previously Zuper, field service · Built the roofing vertical from 0 to 1 · Chennai
Current state
Shipped
In progress
- First 90 days at VoiceStack
- Ask Dilith, site chatbot
Exploring
- Why voice is the right interface for front-office work
- What field service taught me that transfers to healthcare
Most product teams learn about customer problems through tickets, NPS scores, and quarterly reviews.
I spent two years inside the operational layer of one vertical, sitting with contractors, mapping how they actually collect money, watching where workflows break before anyone files a bug report.
What that taught me wasn't roofing. It was that the leverage in vertical software is almost never in the feature list. It's in the operational moment nobody has instrumented yet. It's the moment where a stretched human decides whether the business makes money that day.
In field service, that moment is dispatch. In dental, it's a phone ringing at a front desk that's already short-staffed. I'm at VoiceStack now, doing the same work in a harder domain, where the caller is often in pain, the compliance surface is real, and an AI that loops instead of helping is worse than no AI at all.
It took two years to learn the first vertical. I expect the second to take just as long, and I'd rather say that than pretend otherwise.
Work
Not success stories. Decision logs.
Onboarding: 12–16 weeks → 15 days · 30+ accounts on the golden account template · 6 hours from problem to roadmap decision
Writing
Product teardowns, field notes, and frameworks
from the boundary between operations and product.
Rapido nailed the habit loop. But they're still optimizing for the ideal user. Not the actual one.
Read →
Most imitators copy the format. They miss the emotional architecture underneath.
Read →
Projects
Personal friction,
built into tools.
The Voice Agent
Live on this site. You can talk to it. Latency you feel rather than measure, speech recognition that mangles proper nouns, knowing when the agent should stop talking and get a human. I built it before I had a professional reason to. It's why I had opinions about voice AI before I had experience in it.
SuperPilot
B2B sales reps lose 2 hours a day across 6 deal handoffs. The data to fix it exists. I built the AI layer that synthesizes it into memory.
Iris
I work across four devices. Every day I need to move something between them: a screenshot, a file, a link. Every existing solution added friction I didn't want. So I built one.
Delphi
Foreign founders with Delaware C-Corps face a $25,000 penalty for missing Form 5472. Most don't know it exists. I built the tool that tells them exactly what they owe, before they talk to a CPA.
Stack
The AI PM toolkit.
Models, automation platforms, and product tools I reach for every week.
- Claude
- GPT-4
- Figma
- Notion
- Linear
- Supabase
- n8n
- ElevenLabs
- Deepgram
- Postman
- VAPI
- Lovable