NeighborhoodAI

Hugo Barbosa · Applied AI Builder · San Diego

Understanding people. That’s where my AI work starts.

Fourteen years in hospitality taught me to listen, understand what people need, and see how work actually gets done. I bring that experience to customer tools, AI agents, and content workflows—starting with the people who need to use them.

Explore the work Résumé PDFLet’s talk

Open to roles applying AI in real businesses, and independent projects.

01 / Selected work

Useful tools.
Built around real work.

Explore a live tool, see the code, or read the decisions behind the build.

yannis-ai-buildPYTHON

01 Read the inquiry

02 Check the business rulesDeterministic tools

03 Classify & draft

04 Hand it to a person

166hand-labeled historical inquiries
Development set

System design · Shadow mode · Draft-only writes

01 / FEATURED BUILD · APPLIED AI

Catering inquiry agent

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

A Python agent built to qualify catering inquiries and draft replies. Pricing and service rules stay in explicit tools, with human approval required before sending.

134 / 134Pytest tests passingDevelopment evidence; not held-out accuracy.

What failed / Adoption

A working tool still needs a place in the day.

The team did not adopt the agent. My lesson: meet the existing workflow, solve an immediate bottleneck, and teach people how to use it.

Read the code & build notes
PythonAnthropic APIEvaluation harnessTool calling
How I built it

The problem

An inbox assistant needs to understand service limits, delivery boundaries, and when to ask for help. A plausible answer is not enough.

What I built

An evaluation harness built around 166 hand-labeled historical inquiry threads used as the development set, separate classification and drafting prompts, deterministic business-rule tools, and structured final outputs. Shadow mode and draft-only writes keep the operator in control; the CRM provides the review surface.

Where I drew the boundaries

The model
Classifies the inquiry and drafts a reply.
Explicit tools
Supply pricing and service rules.
The operator
Reviews the draft and approves sending.

The decision that matters

Build the evaluation workflow before trusting the agent. Keep pricing and service facts in explicit tools, and make exceptions visible to the operator.

The 166 labeled inquiries and passing tests are development evidence, not a held-out accuracy result. Draft-only writes and human approval are implemented safeguards. Team adoption did not follow; the lesson was to fit the existing workflow, address an immediate bottleneck, and teach people how to use the tool. A revised rollout has not been demonstrated here.

YANNI’S EVENTSWorkflow overview

Keep the whole
event together.

Booking details
Menus & documents
AI drafts. A person reviews.Approval stays with the operator.
PRIVATE PLATFORM · WORKFLOW OVERVIEW

02 · OPERATIONS SOFTWARE · PRIVATE

Events platform

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

Bookings, menus, and event documents in one private workspace. AI creates drafts for review, and repeated requests are handled without creating duplicate work.

Ask for a walkthrough

The decisionRepeated requests should not create duplicate work.

Next.jsTypeScriptSupabaseAgent integration
How I built it

The problem

Event details need to move from inquiry to an operational plan. Scattered messages make that handoff difficult to follow.

What I built

A protected application for bookings, menus, and event documents. The agent integration supports draft-only writes, explicit human approval, and a shadow mode for observing proposed actions.

Where I drew the boundaries

The agent can propose work through draft-only writes. A person retains approval, access controls keep the workspace private, and repeated requests are handled without creating duplicates.

The decision that matters

Repeated agent requests should not create duplicate work. Idempotency makes a repeated inquiry safe to process, while access controls keep the review surface private.

Deployed, private application. Broader team adoption remains limited; the build includes automated tests, documented operating boundaries, and a human review workflow. Walkthrough available.

YANNI’S BAR & GRILL

What would your
event look like?

YOUR GUESTS40 guests
Menu packageMenu 3
BeveragesBeer & wine
Build your estimate
INTERFACE OVERVIEW · ILLUSTRATIVE SELECTIONS

03 · CUSTOMER TOOL · LIVE

Private event estimator

My contributionThe customer journey, pricing logic, quote delivery, and activity tracking.

A private event estimator that turns guest count, menu choices, and drinks into an itemized budget. A useful starting point before the first conversation.

The decisionPricing stays in explicit rules. Guests get a concrete starting point.

Try the estimator
Next.jsTypeScriptSupabaseResend
How I built it

The problem

Planning an event starts with pricing questions. The restaurant’s rules need to make sense to someone who has never booked there.

What I built

A customer-facing quote workflow with menu and beverage choices, itemized estimates, branded PDFs, email delivery, shareable quotes, and view tracking. Admin analytics make the quote activity visible.

How the pieces fit

  1. Guest count and menu choices
  2. Explicit pricing rules
  3. Itemized estimate
  4. PDF, email, and a shareable quote

View tracking and admin analytics make activity visible after the estimate is shared. The output gives the guest and the team the same budget to discuss.

The decision that matters

Keep the pricing logic explicit. Let the guest explore a budget, then give the team a concrete starting point for the booking conversation.

Additional tooling: PostHog, Playwright, Vercel. Published application; usage tracked, no conversion claim yet.

A bottle detail page in the actual cellar application, including price and availability guidance.ACTUAL INTERFACE · FROM THE WALKTHROUGH

04 · DIGITAL EXPERIENCE · LIVE

Wine cellar

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

A searchable inventory of 275 wines, with tasting notes and food pairings. Guests can inquire about a bottle for a visit or take-home purchase.

The decisionMake discovery easy. Keep availability confirmation with the team.

Explore the cellar
Watch walkthrough 0:28
Silent tour: search and filter wines, explore a bottle’s details and pairings, then find inquiry options for a visit or take-home purchase.
Open video
Next.jsSearch & filtersContent researchWine catalog
How I built it

The problem

A large collection is only useful if a guest can find their way through it. Bottle names alone leave a lot of the story untold.

What I built

Search and filters for section, country, body, and price. Individual wine pages bring together tasting descriptions, producer background, pairing ideas, and inquiry options. I also researched descriptions and built a searchable Airtable wine database.

How the pieces fit

Filters narrow the collection. Bottle pages supply context through tasting notes, producer background, and pairings. Inquiry options connect that discovery to a visit or take-home purchase, with the restaurant confirming availability.

The decision that matters

Connect discovery to hospitality. Guests can inquire about a wine for a visit or take-home purchase, with the restaurant handling the next step.

275 wines visible in the public collection as of September 2, 2026. This is an inquiry experience, with availability confirmed by the team.

A stone-walled vineyard in Mamoiada, Sardinia, from the Wine Adventure journal

06 · ALSO BUILT

Sardinia wine journal: travel photos and stories shaped into a responsive editorial website.

Take the wine adventure

02 / How I work

The business comes first.
The build follows.

The same approach connects the newsletter, the customer tools, and the agent.

01

Find the actual job.

Talk to the people doing the work. Map the rules, the handoffs, and the part that keeps getting stuck.

02

Make it usable.

Build around real inputs and clear decisions. Keep a person involved where judgment matters.

03

Learn from the work.

Test against real cases. Look at what people use. Keep the lessons so the next version starts smarter.

03 / A little context

Hospitality was my education in how people work.

LinkedInGitHub

Hospitality taught me to pay attention: to what someone asks for, what they actually need, and how a team works when the day gets busy. That experience shapes the questions I ask before I build.

The newsletter review taught me to look beyond clicks when evaluating business results. The catering agent went unused because I hadn’t secured team adoption. Both experiences shape how I build: understand the people, choose the right measures, and learn from what happens.

I’m looking for work where I can apply AI to business operations and help turn that work into revenue. I’m open to different role titles and independent projects; the work itself is what interests me.

Understand the people. Build from there.

OPEN TO ROLES & INDEPENDENT PROJECTS

Let’s make AI
useful to your business.

Building a team or working through a business problem? Tell me what you want AI to help with, and who needs it to work.

Start a conversation Résumé PDFhello@neighborhoodai.ai