Proficient Stack Start a build
/ AI Engineering

Production AI features. Not demos.

Most "AI" ships as a prototype that impresses in a meeting and breaks in production. I build the version that lives in your product: RAG chatbots, LLM workflows and agents on the Claude and OpenAI APIs — grounded, evaluated, and maintained. Full-stack in Next.js and Node, remote, in writing, no calls.

Nexa
A live AI receptionist I built & run — multilingual, replies in <1s
115 endpoints
A production platform I operate — payments, subscriptions, webhooks
Claude + OpenAI
I build the agent layer, not just wrap a model

A demo answers once.
A product answers every time — and knows when not to.

# the part that separates a demo from a feature
user → retrieve (RAG, your data — grounded, cited)
    → LLM (Claude / OpenAI, tools + guardrails)
    → confidence low? hand to a human # no confident nonsense
    → log + eval # precision/recall, not vibes
Build
/WHAT I BUILDReal features inside real products — not a chatbot bolted on the corner of a page.
01

RAG chatbots & assistants

Answers grounded in your own data — docs, tickets, catalogue — with citations, so it doesn't invent. Built to hold context and hand off to a human when unsure.

02

LLM workflows & automation

Intake, triage, summarisation, drafting, classification — the repetitive judgement work, automated with the model in the loop and a clean fallback when it isn't sure.

03

Agentic features

Agents that take actions against your APIs and tools — not a toy loop, but bounded steps with retries, idempotency and an audit trail you can trust.

04

AI inside your existing app

You have a product; you want an AI feature in it. I wire it into your Next.js/Node stack and ship it as part of the app, deployed and maintained.

05

Grounding & guardrails

The unglamorous half: retrieval that's actually relevant, prompt/tool guardrails, PII handling, and evaluation on precision/recall so quality is measured, not hoped.

06

UAT → production

Have a prototype stuck in "almost"? I stabilise and scale it — performance, cost, observability — into something you can put real users on.

How
/HOW I WORKBuilt to be maintainable and honest about what AI can and can't do.
·

End to end

Frontend (Next.js/React), backend (Node/NestJS, Python), the model layer and the deploy. One person who owns the whole feature, not a hand-off chain.

·

In writing, no calls

Specs, PRs and async status instead of meetings — fast, traceable, and it fits across time zones. You always see progress.

·

Clean invoicing

Invoiced from Portugal with a VIES VAT number — a clean reverse-charge arrangement for EU clients, straightforward for everyone else.

Service
/DONE FOR YOUFixed scope, fixed price, by text. Under your brand if you're an agency.
Build · ship

An AI feature, in production

Pick the workflow. I design it, build it into your stack, and ship it — grounded, evaluated, and maintainable, not a demo you have to babysit.

  • RAG / assistant / workflow / agent — your call
  • Claude & OpenAI, on Next.js / Node / Python
  • Guardrails, grounding, and eval built in
  • Deployed, documented — white-label optional
Rescue · scale

Get the prototype to production

An AI demo that impresses but isn't reliable enough to launch. I make it production-grade — the reason it's stuck is usually retrieval, guardrails or cost.

  • Fix grounding so it stops inventing
  • Add evaluation, observability and fallbacks
  • Cut latency and token cost
  • Scale the backend for real traffic
Available for AI builds

Tell me the workflow
you'd automate.

Describe the feature or the workflow that eats your team's time. I'll come back, in writing, with how I'd build it — the model, the grounding, and where I'd put the guardrails.

Email me