AI Engineer · Mendoza, Argentina
I build AI agents and software that solve real business problems.
I'm Federico Correa, an AI engineer. I design virtual assistants, n8n automations and RAG systems, and I build the web, mobile and desktop apps around them.
This building is my portfolio: every floor is a project, with its agents working inside. Pick a floor to go in.
7projects, one floor each
4AI and RAG systems
3platforms: web, Android and PC
828% ROIestimated for my agentic project
Floor 7 · Agentic system Final project
Supervised agentic assistant for a wine shop
Case study · Final project, Advanced AI Automation, Coderhouse 2026
The agents on this floor
Handles WhatsApp questions about prices, stock, food pairings, shipping and complaints. A Manager agent classifies each message and routes it to a specialist. No reply reaches the customer without passing an AI supervisor that scores it from 1 to 5, corrects it or escalates it to a person.
WhatsApp→Manager→Sales · After-sales · Complaints→AI judge→Dispatch + CRM
- 8 independent n8n workflows with a Manager-Worker architecture.
- Answers only from indexed information (Pinecone), with cited sources.
- Idempotency, bounded retries, a traceId per message and error alerts.
- HubSpot integration: contacts, complaint tickets and sales deals.
USD 309estimated monthly savings
77 hhuman hours saved per month
1.3 monthspayback period
USD 5.68AI cost per 1,000 messages
Base scenario of 1,000 messages per month, versus USD 417 for fully manual support.
n8nAI-as-a-JudgeRAGPineconeWhatsAppHubSpotAirtableGmail
Floor 6 · CRM integration In progress
Wine shop agent connected to HubSpot
Personal project
The agents on this floor
An extension of the wine shop agent so every customer, complaint and sale is recorded in a CRM: contacts by WhatsApp number, tickets with a status pipeline, sales deals and notes with the supervisor's evaluation.
n8nHubSpotAirtableGoogle Sheets
Floor 5 · Virtual assistant In production
Ceti, an AI sales assistant
For a software company
The agents on this floor
A web and WhatsApp assistant that answers questions about modules and plans, captures lead details, recommends a plan, flags large companies to the sales team and books demos. An AI judge reviews every reply and a fixer corrects the ones it rejects.
- Two backend versions: with Airtable and with HubSpot as the CRM.
- Prompt caching to lower the cost per conversation.
- A Python workflow generator and automated conversation tests.
n8nHubSpotAirtableWhatsAppPythonCloudflare
Floor 4 · Web app + AI Working
CentralVet, veterinary clinic management
App for a veterinary clinic
The agents on this floor
Clients, pets and full medical records: weight, allergies, chronic conditions, medication, vaccines and visit notes. It includes an AI assistant that reads the pet's record to answer questions such as the right dose for its weight and conditions.
- Dose calculator using the clinic's own editable drug table.
- Conversations saved per pet, with streaming answers.
Next.js 16ReactTailwindSQLiteClaude API
Floor 3 · RAG chatbot In progress
Process safety chatbot
Personal project · chemicals, oil & gas, energy, mining and more
The agents on this floor
Answers process safety questions over technical documents, with source citations and live tool status. Ready to query public incident databases.
- Streaming chat (SSE) with citations, on Postgres + pgvector.
- Per-IP rate limits, protected document uploads and prompt injection defenses.
- Backend and frontend in Docker containers, ready for a VPS.
FastAPINext.jsClaude APIVoyage AIpgvectorSupabaseDocker
Floor 2 · API + RAG Working
Process safety library
Personal project
The agents on this floor
A database and API of technical documents: CSB accident reports, standards and procedures (OSHA, CCPS and operators), and HAZOP and LOPA studies, classified by the CCPS's 20 RBPS elements.
- 33 documents and their relationships, modeled with SQLAlchemy.
- Hybrid search (text + embeddings) and LLM-generated answers.
- Web interface, Swagger docs and deployment with Docker and Nginx.
PythonFastAPIPostgreSQLpgvectorGroqOllamaDocker
Floor 1 · Android + desktop Version 1.4
Field route app and admin panel
For a gas distribution company
The agents on this floor
An Android app for field staff, with photo check-ins and route closing that notifies the admin. It comes with a Windows desktop app where admins manage users and routes.
- Supabase backend with row-level security (RLS) policies.
- Desktop app with its own MSI installer.
KotlinJetpack ComposeCompose DesktopSupabasePostgreSQL
What I do
From prototype to something that runs in production
I focus on making AI do a specific job, with real business data, quality checks and measured costs.
AI assistants and agents
Web and WhatsApp chatbots that qualify leads, answer with the business's own information and hand off to a person when needed.
n8n automations
Workflows connecting CRMs (HubSpot), Airtable, Gmail, Google Sheets and WhatsApp, with retries, alerts and traceability.
RAG over documents
Semantic and hybrid search over manuals, standards and knowledge bases, with source citations and a rule to say "I don't know".
Web, mobile and desktop apps
Next.js, FastAPI, Kotlin and Compose. PostgreSQL, Supabase and SQLite databases, packaged with Docker.
About me
Technology applied to concrete problems
I work in AI automation and full-stack development. I like taking a process that's done by hand today, like answering inquiries, logging customers or searching documents, and turning it into a reliable system.
I'm currently an AI Automation Specialist at Grupo CETAP. I completed the Advanced AI Automation program at Coderhouse (2026), where I designed a supervised agentic system with a cost and ROI analysis. I have a particular interest in industrial process safety: several of my projects apply AI to standards, HAZOP and LOPA studies and incident reports. I work in Spanish and English, and I'm open to remote roles.
How I work
- Understand the problem and what it costs today in time and errors.
- Build a quick prototype with real data to validate the idea.
- Add quality checks: cited sources, supervision and handoff to people.
- Measure costs and document everything so the team can maintain it.
What I work with
Claude APIGroqOllamaRAGEmbeddingsPineconepgvectorPrompt cachingn8nHubSpotAirtableWhatsAppPythonFastAPITypeScriptNext.jsReactKotlinPostgreSQLSupabaseDockerCloudflare
Contact
Hiring an AI Engineer, or have a process to automate?
Tell me what you need, whether it's a role on your team or a project, and let's see if an AI agent or a custom app is the right fit.