Left: the backoffice, where the owner teaches Hugo. Right: what the customer sees on WhatsApp.
"Every week, millions of customer messages get lost. Who's actually taking care of your customers when you're not there?"
— the hook, straight off the landing page
Overview & context
Customer success shouldn't need a headcount.
Most Customer Success tooling promises the world at the sales call, then leaves a gap between onboarding and real implementation — weeks of setup most small businesses never finish.
Meanwhile customers expect an answer almost immediately, on the channel they already use. When they don't get it, they don't complain — they just leave.
I designed the product and turned the team's idea into flows and screens. Jerónimo, Ivo and Máximo vibe coded it during one hackathon, from problem to a live WhatsApp agent handling real conversations.
55%
of US shoppers won't return to a brand after one bad experience
75%
of online customers expect help within five minutes
A small business should be able to trust an agent with real customers from day one, with no implementation project.
Design
My part
Code
Jerónimo, Ivo and Máximo
Process · Vibe coded, start to finish
How one night was sequenced
Five stages
The loop had to work before any interface existed.
Vibe coding doesn't leave room for exploration — it forces sequencing. The loop had to work before the interface existed. Here's how it actually broke down.
Stage 01
Framing the problem
Chose AI Customer Success because response speed is retention, not a feature on a backlog. The brief: an agent a business can trust with real customers, on day one.
Stage 02
The loop, before any UI
Designed the ask → answer → teach → memorize cycle first, as a state machine. If the loop didn't hold, no interface would save it.
Stage 03
WhatsApp as the only channel
No new app to download, no dashboard to log into. Hugo meets the customer exactly where they already are.
Stage 04
Editable memory
Gave the business owner a direct way to correct Hugo mid-conversation, in plain language — no prompt engineering required.
Stage 05
A live demo, not a pitch deck
Recorded a 60-second demo of the real product answering real questions. A working loop is a stronger pitch than any slide.
Core mechanic · The learning loop
Every correction becomes permanent memory.
Hugo isn't smart because of a bigger model. Its real asset isn't the answer engine, it's the loop that keeps improving it.
Step 01
Customer asks a question
Straight on WhatsApp, in their own words. No ticket, no form, no waiting for business hours.
Step 02
Hugo answers
Using the business's existing memory — real data, not a generic script.
Step 03
If Hugo doesn't know, it asks the owner
Instead of guessing or stalling, Hugo escalates the gap directly and honestly.
Step 04
Owner corrects / teaches Hugo
In plain language — typed in the backoffice chat, or spoken directly to Hugo on a live call. No prompt engineering either way.
Step 05
Hugo memorizes and improves
The correction becomes part of permanent memory, not a one-off patch.
Step 06
That knowledge becomes the system
Every future customer conversation benefits from what was just taught.
The core principle
"Each company trains its own Hugo. We train the ecosystem."
Every conversation is a chance to learn, and every correction makes every Hugo better — not just the one that was corrected. That compounding loop, not the model underneath, is the actual moat.
Straight from the hackathon deck — same slide, same night
Or just call Hugo — teaching it by voice, live
Product · How it works
How Hugo works, in four parts.
01 · Onboarding
Instant onboarding
Setup in minutes through a ChatGPT-style interface — no implementation team, no sales call, no waiting.
02 · Channel
WhatsApp-native
Connects directly to the channel customers already use — answers in text or a real voice note, whichever fits the moment.
03 · Response
Automated, current answers
AI responses backed by real, up-to-date business data — not a static script that goes stale.
04 · Memory
Editable memory
Learns from every interaction. The owner can correct it by typing — or just calling Hugo and saying it out loud.
Beyond the demo · The plan we pitched
First use case: event ticketing.
Hugo's first real use case was WhatsApp support for event ticketing — a category with a brutal spike-and-silence pattern that makes response speed non-negotiable.
But the loop underneath (ask → answer → teach → memorize) isn't ticketing-specific. It generalizes to any SMB drowning in repetitive, time-sensitive questions — e-commerce, courses, healthcare came next in the roadmap, sketched out the same night.
$25→$299+
Proposed monthly pricing. Usage and add-ons scale the plan
0
Implementation needed — product-led, self-serve from day one
2
Buyer segments mapped: ticketing founders first, then e-commerce/courses/health
Market, as pitched
500K+
event organizers and SMBs across LATAM. This is the estimate from our pitch, not a measured figure.
Go-to-market, sketched the same night
Product-led, through WhatsApp itself.
Acquisition leans on product-led growth and the WhatsApp integration itself as the entry channel — a live demo is two clicks away. From there: alliances with ticketing, e-commerce and payments platforms, plus word-of-mouth from the organizers already running Hugo.
PLG
Primary channel
Low
CAC via inbound + content
Min.
Support needed to scale
Results · What we shipped that night
A product that didn't exist yesterday.
Hugo was built in one night, so there are no usage metrics yet. What we can show is that it works.
Live
A WhatsApp agent answering real conversations, open for anyone to try
60 s
Demo of the real product, recorded the same night. Not a mockup
2
Ways to teach Hugo, both working: typing in the backoffice or calling it
Learnings
Learning 01
One night forced the order of the work
Vibe coding forced the loop to be designed before the interface — and the product was better for it.
Learning 02
A working chat beat a polished dashboard
The demo was the product answering on WhatsApp, not screens of what it would look like.
Learning 03
Memory is the real product
The UI is disposable. What compounds — and what's defensible — is what Hugo remembers about each business.
The team
Built with Jerónimo, Ivo and Máximo.
I designed the product and shaped the team's idea. Jerónimo, Ivo and Máximo built it in one hackathon, using from021.io and v0: Next.js, Tailwind and shadcn/ui, scaffolded and shipped the same night.
