NeighborhoodAI

The work behind Neighborhood AI · Hugo Barbosa

The problem. The build.
What happened next.

A closer look at six projects from my work at Yanni’s Bar & Grill: the need, how I approached it, and what we know so far.

Choose a project

Project case studies

Public tools, private builds, and lessons from use.

01 / Marketing

Newsletter & content workflow

In use at Yanni’s

  1. Business background
  2. This week’s story
  3. Draft, edit & send
  4. Review & learn

My partBusiness background and brand voice, the production workflow, newsletter writing and builds, repurposing, and performance review.

Read the newsletter archive
The problem
The restaurant has specials, stories, and events to share. Getting a good email out repeatedly means keeping the facts, voice, and current offer together.
What I built
ServePost: a repeatable process that carries business context, approved examples, and previous edits into the next newsletter.
How it addresses the problem
Start with a real reason to visit, draft from the business’s background, edit and check the finished email, then use what we learn in the next cycle.
What we know so far
28 measured sends, February 18–August 26, 2026. The later 11 averaged 46.4% opens. This is repeated use in one restaurant; opens alone do not establish additional business.
Build details

What the review changed

Galleries and videos drew stronger click engagement than booking or review requests in the later sample. That informed a content-led main button with business asks kept secondary. These are observational comparisons, not controlled tests.

Engagement and sales answer different questions

Toast attributes approximately $95,766 and 731 orders across March–August 2026. May is the sum of four campaign-page figures; the other months use monthly summaries. This is platform attribution, not independently measured incremental revenue or revenue generated by ServePost.

What carries forward

Business background, voice, approved content and useful corrections stay in readable documents. Human review remains part of every send. The review loop records the question before drafting and aggregates results seven days after the actual send.

Tools: ServePost · business documents · AI-assisted drafting · Toast

Back to all projects ↑

02 / Customer information

Private event estimator

Live customer tool

  1. Guest count & menu
  2. Explicit pricing rules
  3. Itemized estimate
  4. Booking conversation

My partThe customer journey, pricing rules, quote delivery, and activity tracking.

Try the event estimator
The problem
A guest planning an event needs to understand the likely cost before working through every detail with the restaurant.
What I built
A customer-facing estimator that turns group size, menu choices, and drinks into an itemized starting budget.
How it addresses the problem
Guests explore their choices against explicit pricing rules. The estimate gives the guest and the restaurant the same breakdown to discuss.
What we know so far
The estimator is published, with quote activity tracking available. A change in inquiries, booking conversion, or staff time has not been established.
Build details

From estimate to conversation

Menu and beverage choices feed an itemized estimate, with branded PDFs, email delivery and shareable quotes. View tracking and admin analytics make quote activity visible.

The key decision

Keep pricing in explicit rules. The tool provides a starting point for the booking conversation; the team handles the details that require a person.

Tools: Next.js · TypeScript · Supabase · Resend · PostHog · Playwright · Vercel

Back to all projects ↑

03 / Product discovery

Wine cellar

Live catalog

Actual wine cellar bottle page with details, price, and availability guidance.
Actual interface · from the walkthrough

My partWine research, searchable catalog, bottle pages, and guest inquiry paths.

Browse the wine cellar
The problem
A large wine collection is hard to explore from bottle names alone. Guests need help finding something that suits their tastes and meal.
What I built
A searchable catalog with filters, bottle details, tasting descriptions, producer background, pairings, and inquiry options.
How it addresses the problem
Filters narrow the selection; bottle pages explain the choices. Guests can then ask the team about a bottle for a visit or take-home purchase.
What we know so far
275 wines were visible in the public collection on September 2, 2026. This shows the catalog’s scope, not a sales result. The restaurant confirms availability.
Build details

Helping a guest choose

Search and filters cover section, country, body and price. Bottle pages connect descriptions, producer background and pairings to inquiry options. A separate researched Airtable database supports the wine work; automatic synchronization is not claimed.

Watch the cellar walkthrough 0:28
Silent tour: search and filter wines, explore bottle details and pairings, then find inquiry options for a visit or take-home purchase.
Open video

Tools: Next.js · search and filters · wine research · Airtable

Back to all projects ↑

04 / Inquiry handling

Catering inquiry agent

Development build · Not adopted by the team

  1. Read the inquiry
  2. Check business rules
  3. Classify & draft
  4. Human review

My partClassification and drafting workflows, explicit business-rule tools, and the evaluation harness.

Read the code & build notes
The problem
Replying to catering inquiries requires understanding the request, checking service and pricing rules, and recognizing when a person needs to step in.
What I built
A Python agent that classifies inquiries and drafts replies, using explicit tools for business rules and leaving approval with the operator.
How it addresses the problem
The model interprets and drafts. Tools supply service and pricing facts. Shadow mode and draft-only writes let a person review proposed work before sending.
What we know so far
The intended team did not adopt it. Recorded development evidence includes 166 labeled historical inquiries and 134 passing tests; neither is a held-out accuracy result. A successful revised rollout has not been demonstrated.
Build details

The adoption lesson

A working tool still needs a place in the day. The lesson was to fit the existing workflow, address an immediate bottleneck, and teach people how to use it.

Evaluation and boundaries

The 166 hand-labeled historical threads were a development set. Classification and drafting prompts, deterministic business-rule tools and structured outputs support review in the CRM. The passing tests are recorded development checks, not a fresh test run for this case study.

Tools: Python · Anthropic API · tool calling · evaluation harness · Pytest

Back to all projects ↑

05 / Event operations

Events platform

Private application · Limited team adoption

My partThe booking workspace, menu and document workflows, and agent integration.

Ask for a private walkthrough
The problem
Event details need to become an operational plan. Keeping bookings, menus, and documents across scattered messages makes that handoff hard to follow.
What I built
A protected workspace that brings bookings, menus, and event documents together, with AI-generated drafts for review.
How it addresses the problem
The workspace keeps related information together. Access controls protect it, a person approves proposed work, and repeated requests are handled without creating duplicates.
What we know so far
The application is deployed privately. Broader team adoption remains limited. Automated checks and documented review boundaries support the build; operational improvements have not been quantified.
Build details

The key decision

Repeated requests should not create duplicate work. Idempotency makes repeat processing safe, while draft-only writes, shadow mode and explicit approval keep the operator in control.

Tools: Next.js · TypeScript · Supabase · agent integration

Back to all projects ↑

06 / Storytelling

Sardinia wine journal

Published editorial website

A stone-walled vineyard in Mamoiada, Sardinia, from the wine journal.
From the Sardinia wine journal

My partShaping the travel photos and stories into a responsive website connected to the restaurant’s newsletter.

Read the wine journal
The problem
The restaurant had travel photos and wine stories to share. They needed a place where readers could explore the trip beyond a short email.
What I built
An editorial website bringing the Sardinia wine journey’s photos and stories together.
How it addresses the problem
The newsletter provides an entry point. The journal gives interested readers room to explore the places, people, and wine behind the story.
What we know so far
The journal is published and linked to the newsletter work. A business impact attributable to the journal has not been measured.
Build details

The content decision

Give the story its own space to be read and explored. Photos and travel detail carry the experience, with a responsive layout for readers arriving from an email.

Work: editorial structure · photo selection · responsive website · newsletter connection

Back to all projects ↑

About Hugo

Understanding people.
That’s where the work starts.

LinkedInGitHub

Fourteen years in hospitality taught me to listen, understand what people need, and see how a team works when the day gets busy. That experience shapes the questions I ask before I build.

The newsletter taught me to separate engagement from business results. The catering agent taught me that a working tool can still go unused. Both inform how I approach the next project.

I take independent projects and am also open to roles applying AI in business operations.

Résumé PDF

Start with your business

Have a problem
you’d like to work through?

Tell me what happens now, where it gets difficult, and what you’d like to change.

Email Hugo hello@neighborhoodai.ai