I build AI systems that real businesses run on, every single day.
Production chatbots, automation pipelines, and dashboards working right now inside logistics, events, medical, manufacturing, and procurement companies. Below you can try one, see how they're built, and check the numbers.
This is what your customers would talk to.
A working simulation of an AI receptionist I build for service businesses. Left: what the customer sees. Right: what the system actually does on each message.
Luna Studio is fictional — the flows are real. In a deployment, these steps write to your actual calendar, CRM, and WhatsApp Business account, answer from your own price list and FAQs, and hand off to a human with full context when needed.
Seven real builds, with the numbers they produced.
Client names withheld out of respect for their privacy. Every metric comes from delivery documents and status reports, and I'm happy to walk through any build in detail.
The problem
A freight forwarder (refrigerated exports to Japan and Europe) ran on manual container tracking, hand-captured documents, and know-how living in one senior operator's head.
What I built
A conversational agent the team chats with all day: it answers operations questions, tracks carrier containers automatically, runs financial analysis, and searches company SOPs through a vector knowledge base. Eight coordinated workflows, including a 7 a.m. daily summary and a weekly financial report.
How it flows
Stack
n8n (self-hosted) · OpenAI · Supabase (Postgres + vector) · Google Workspace
The problem
A corporate events agency (its clients include DHL, Tetra Pak, and Mazda) spent hours per proposal reading briefs, classifying requirements into 12 cost categories, and hunting suppliers by hand.
What I built
A pipeline that watches for new brief PDFs, extracts and classifies every requirement with structured LLM outputs, audits its own work with a second model, finds and verifies suppliers, and contacts them via WhatsApp and email.
How it flows
Stack
n8n · OpenAI structured outputs · Supabase · SerpAPI
The problem
A medical group's monthly marketing report took days of manual consolidation across Meta Ads, Google Ads, and lead forms, while budget decisions were made without current data.
What I built
A secured dashboard syncing both ad platforms daily with failover, joining spend, leads, and revenue per treatment line: funnel views, year-over-year comparisons, investment-versus-revenue crosses, prioritized insight alerts, and PDF exports.
How it flows
Stack
Ad platform APIs · automated ETL · hardened web dashboard
The problem
The global procurement division of a multinational manufacturer assembled its monthly raw-materials report by hand: index prices copied from exports and PDFs, reconciled against paid prices, deck built manually.
What I built
A Python pipeline ingesting market sources automatically, reconciling against paid prices, and generating a password-protected dashboard with an AI-written quarterly thesis, market-versus-paid gap flags, a natural-language "ask the data" box, and one-click deck export.
How it flows
Stack
Python · Claude API · automated Excel/PDF ingestion · PowerPoint export
The problem
A high-end custom carpentry studio quoted from a fragile spreadsheet with 485 broken references. The owner's fear: digitizing would lose the craft knowledge inside those files.
What I built
I mined their 81 GB archive (719 spreadsheets, 640 quotations) into 814 clean material SKUs, reconstructed their real pricing logic from 5,000 cross-sheet formula references, and built a quote-by-spaces web app with a living price catalog, automatic markups, and AI suggestions from similar past projects.
How it flows
Stack
Python analysis · Supabase · web quoting application
The problem
Executives learn AI fastest when the material is their own work. This leadership team wanted AI fluency applied directly to supplier negotiations, commodity analysis, and market reports.
What I built
A ten-module hands-on program built on their own documents and data, with a practical framework for deciding what to delegate to AI and how to verify its output. Each director leaves with a configured AI assistant and a personal prompt playbook. I run the same format in English and Spanish.
Format
Why it exists
I wanted a complete marketing employee: strategy, calendar, copy, and execution across multiple brands, all coordinated in one system. So I built her.
What it is
A multi-agent system — eight specialized agents including a content analyzer and an advanced copywriter — that turns brand strategy into calendars, posts, and ad creative with anti-repetition logic, connected to Meta and Google Ads.
Stack
FastAPI · React · PostgreSQL · Redis · MinIO · Docker
The actual workflows behind the demos.
These are real n8n workflows I designed — importable files, not diagrams. Every demo above maps to one of these builds.
End to end, and honest about it.
I chart your actual flow first, so every piece connects instead of living in its own silo. The diagnostic often pays for itself before any automation ships.
You see a functioning version in the first days, not a slide deck. I'd rather show a rough build than describe a perfect one.
Self-hosted infrastructure, error handling, monitoring, documentation. Systems my clients run daily after I'm gone.
A degree in business and IT, a master's in AI, and 8 years of full-stack development. I talk to owners and to APIs.
Itzel Orozco
I'm a full-stack developer with 8 years of experience — an AI developer for the last 5, and focused on LLM systems for the past 3. With a degree in business and IT and a master's in AI, I move between the business side and the technical side without a translator.
I work end to end: strategy, architecture, development, and deployment to production. Fluent in English and Spanish, working with clients across Europe, North America, and Latin America.
Have a system in mind?
Message me on Upwork and tell me what's eating your team's time. I'll usually answer with a sketch of how I'd build it — and often with a small working prototype.
Message me on Upwork