AI without the theatre.
Production systems that use LLMs, retrieval and agents where the return is measurable. Shipping on OpenAI since the Davinci alpha (January 2021) and on Claude since launch.
Technology that fits your people. Not the other way around.
Proportione builds technology around how your team already works — then transfers the knowledge so your team can run it without us.
206020
Our operating model
Most projects we are asked to rescue have one thing in common. The technology works. The team does not use it.
Before anything else, we look at how your people already work. Then we build around them — not the other way.
How we work
20 · 60 · 20.
20%
Your people decide.
Strategy, constraints, what matters and what does not. Your team frames the problem.
60%
AI does the heavy lifting.
Code, drafts, transformations, retrieval. The repetitive, the dense, the slow.
20%
Your people validate.
They check, they own the decision, they sign off. The accountability stays with them.
Humans stay in charge. Technology does the heavy work.
What we are hired for
Platforms that hand back the keys.
Cloud-native platforms — CRMs, learning environments, internal portals. Built so your team can run them without us when the time comes.
Automation as engineering.
Pipelines that connect CRM, marketing, support and operations. We treat automation as code, not as plumbing.
Security from day one.
GDPR by design. Secrets in Secret Manager. Audit trails on by default. We document what we ship.
Selected work
- Osakidetza — Curated catalogue of public-health video content for a regional learning environment.
- IITD — Eighteen-month operating partnership: platform, secretariat panel, certificates, CRM.
- porqueViven — Public open-source repository: QR system for BIM-tagged elements.
What we publish
Engineering insights.
-
Servidor propio para una consultoría AI-first: cuatro opciones, dos trampas y una decisión
Decidimos no migrar a un hyperscaler ni encadenar Macs. Va el marco con el que elegimos servidor propio, la BOM real, las dos trampas que casi rompen la...
-
Capataz y Obrero: por qué Claude Code delega a un LLM local en lugar de hacerlo él
Claude Code para todo es caro y lento. Un modelo local para todo es limitado. La combinación correcta no es elegir uno: es dividir el trabajo. Va el...
-
Prospección B2B industrial sin email frío: cuatro herramientas y una regla
El email frío ya no rinde. El uno-a-uno manual tampoco escala. Este es el pipeline que construimos para prospección industrial en Portugal — cuatro piezas, dos decisiones de...
-
Cuando el router inteligente desconecta tus IoT: el caso del Roaming Assistant
Seis Alexas dejaron de responder a las 22:00 y no volvieron solas. La causa no estaba en las Alexas ni en Amazon — estaba en un parámetro del...
Next step