Iberian AI consultancy · Headquartered in Aveiro proportione.com · engineering.proportione.com · voxelers.com · LinkedIn
Most AI initiatives fail at the people layer, not the model layer. We were shipping recommender systems before machine learning was a board-level acronym, training teams on GPT-3 twenty months before ChatGPT, and operating production AI agents today. Proportione, founded in 2021, is what twenty years of that work made possible.
Our practice is built around three vectors: strategy — where AI actually creates leverage; research — applied work, publications, and a doctoral programme at the Universidade de Aveiro; and engineering discipline — the controls that turn pilots into production.
flowchart LR
A["<b>20%</b><br/>Define<br/><sub>your team frames<br/>the problem</sub>"] --> B["<b>60%</b><br/>Execute<br/><sub>AI handles the<br/>routine work</sub>"] --> C["<b>20%</b><br/>Validate<br/><sub>your team learns,<br/>refines, owns</sub>"]
style A fill:#F5EFE6,stroke:#5F322F,stroke-width:2px,color:#5F322F
style B fill:#5F322F,stroke:#5F322F,stroke-width:2px,color:#F5EFE6
style C fill:#F5EFE6,stroke:#5F322F,stroke-width:2px,color:#5F322F
Seventy percent of transformations fail because of people, not technology. The 20·60·20 framework keeps human intelligence at the boundaries of every workflow — defining what matters and validating what works — while AI takes the routine middle. It is the only configuration we have seen consistently produce sticky outcomes.
Aviaria Civil — Bird-strike risk for civil aviation
Real-time bird-hazard intelligence for pilots and small aerodromes. Twenty-two risk layers (eBird, GBIF, migration corridors, power lines, terrain), route analysis over CORINE land cover, PWA installable on iPad in the cockpit, offline-ready. Built in collaboration with Spain's BACSI (Air & Space Force bird-strike committee). Free for civil aviation operators.
A Python toolkit for systematic literature reviews. PRISMA 2020 reporting, MMAT 2018 quality assessment, full-text extraction, bibliometric clustering. CLI + Streamlit demo. Born from our doctoral research at UA.
MIT · research · prisma-2020 · python
Doctoral research at Universidade de Aveiro in Business Innovation, working at the intersection of generative AI and organisational change. Field notes on the engineering blog.
PhD · research
Our internal Claude Code marketplace, opened to clients. Nine quality skills — deep review, security, architecture, performance, WordPress, Terraform — and eight playbooks distilled from production work.
MIT · claude-code · mcp
Configurations, hooks, slash commands, and MCP patterns we use daily. Reusable material for teams bringing agentic AI into production.
MIT · agentic-ai · showcase
Voxel-native 3D stack — Blender, MagicaVoxel, Godot, procedural generation. The geometric primitive behind our digital-twin and creative work. See voxelers.com.
Apache-2.0 · digital-twin · voxels
Engineering blog. Post-mortems, architecture decisions, and applied research from the team — written in English, for technical audiences.
Jekyll · publications
Open-source work with the porqueViven foundation. Tooling for caregivers of chronically ill children — pro-bono engineering, public license.
MIT · third-sector
Podcast em português sobre IA, indústria e ecossistemas ibéricos. Episódio 01: dois anos em Aveiro, em silêncio. Spotify · YouTube
pt-PT · podcast · Aveiro
Every change is peer-reviewed under branch protection. Production deploys require manual approval in a separate environment. Every decision is in the audit log. Our clients do not depend on a person — they depend on a process that can be inspected, reproduced, and where appropriate, opened.
Issues · Discussions · Security policy
Aveiro, Portugal — Rua D. Jorge de Lencastre 10A Madrid presence · Iberian footprint · LATAM-ready