Open to Applied AI roles · Ireland or remote

Augusto Bastos

Applied AI Engineer building RAG, MCP and the tested backends behind them.

Python · TypeScript · LLM/RAG · MCP · AI Automation · Limerick, Ireland

RAG over pgvector MCP server for AI agents Postgres RLS tenant isolation 18 test vectors, 3 languages

View projects Download CV

Based in Limerick, Ireland · open to roles in Ireland or remote

Work

Featured Projects

Four systems I designed and built, each with a link to its code, a running build or a write-up.

Flagship · Applied AI · RAG

pgvector retrieval Grounded answers Open-source rag-demo

PitchPilot & rag-demo: Retrieval-Augmented Generation

PitchPilot is a copilot for field sales reps: a TypeScript edge function embeds the question with Gemini, searches pgvector inside the rep's tenant and grounds the answer in what it retrieved, re-embedding only the chunks whose content changed. When an IVFFlat index started returning nothing on a few hundred rows, I replaced it with exact search. rag-demo is a small open-source Python version of the same idea: overlapping chunks, local embeddings, pgvector and a Gemini answer that cites the files it used, with offline tests in CI.

TypeScript pgvector Gemini Python Docker

Read the case study → rag-demo on GitHub →

Local-first systems · Open source

Wavr: Explainable Presence Sensing with an MCP Interface

A Python and FastAPI engine that combines network, Bluetooth and camera evidence into an explainable per-room state: each source is weighted by trust and decayed with age into a confidence, so it can say whether anyone is in a room, how sure it is and why. AI agents query it over MCP (stdio and HTTP; HTTP off by default, the one write tool stdio-only and opt-in). It binds to loopback by default and never stores camera frames. CI runs the regression suite plus mutation checks that delete each guarantee on its list and fail unless a test notices. AGPL, used by one person so far.

Python + FastAPI Sensor Fusion MCP server pytest + CI

Read the case study → View on GitHub →

Product engineering · Pre-launch

Lucky Cat: Multi-Tenant Restaurant SaaS

One multi-tenant Supabase backend for customer ordering, restaurant operations and AI-assisted sales tooling. Row-level security isolates each tenant's data in PostgreSQL, checked by Playwright tests that try to read, update and insert across tenants. Payments use Stripe Connect direct charges with application fees, and stock deduction runs as an idempotent database function. Pre-launch: no customers yet.

Ownly Owner dashboard: shifts, payroll, stock, cash-up Tillr Customer ordering app in React and TypeScript

Supabase + RLS React + TypeScript Stripe Connect Cloudflare Pages

See the platform →

Protocol Design · Open Source · Frozen at v0.2.3

SCPE: Verifiable Provenance for Code Contributions

A specification for checking, offline, which git-host account signed a code contribution, using the SSH keys that host already publishes. I wrote the spec and a threat model; Python, Go and Rust implementations must return the same verdict on 18 normative test vectors, plus a differential test on mutated inputs. It shipped on PyPI and as a GitHub Action. When I asked the maintainers it was designed for how they check AI-use policies, their answers disproved the premise, so I froze it and put that finding at the top of the README.

Protocol + Spec Python · Go · Rust SSH signatures GitHub Actions

View on GitHub → PyPI →

Also open source

github.com/augbastos The source for Wavr, rag-demo and SCPE, plus smaller public projects. View on GitHub →

Stack

Technical Skills

The stack behind the projects above.

RAG / Vector Search

MCP Servers

LLM APIs & Grounding

Python + FastAPI

Supabase + Postgres + RLS

Testing & CI

Stripe Connect

Cloudflare Pages & Workers

Background

About

Applied AI Engineer based in Limerick, Ireland. I build AI software in Python and TypeScript: retrieval pipelines over pgvector, MCP interfaces, and the backends, data isolation and tests they depend on.

I came to software through IT support and teaching: four years as a receptionist and IT technician, then three years teaching English. Today I also work as a Retail Coordinator at Chemist Warehouse, coordinating a shift team in a high-volume pharmacy.

Location Limerick, Ireland

Languages English C2 (Cambridge Proficiency) · Portuguese (Native)

Availability Open to Applied AI and AI software engineering roles, in Ireland or remote

Experience Retail Coordinator, Chemist Warehouse (2026–) · English Instructor, WSA Languages (2019–2022) · Receptionist & IT Technician, Unified Educational Foundation (2014–2018)

Education Graphic Design, Universidade Veiga de Almeida

Contact

Hiring for an AI engineering role?

I'm looking for a full-time role building applied AI software, in Ireland or remote. Email is the fastest way to reach me, and the CV has everything on one page.

Email me Download CV