PV Paulo Vila Architect of Local-First AI Systems
Ask Paulo's AI For Recruiters

Sovereign AI systems, built in Rust — running on your hardware.

For legal, health, insurance and defence teams in the EU that cannot send data to a US cloud. I design, build and deploy on-premise LLM infrastructure — and leave you the running system, not a slide deck.

Book a 30-minute call

What I offer

What I offer

Sovereign AI Audit

5 days
EUR 2,300 / USD 2,500
  • Inventory of where your data leaves the jurisdiction today
  • On-premise feasibility per use case — hardware, models, cost
  • Written 90-day plan with numbers

Fixed price, fixed scope.

Book a 30-minute call

On-premise RAG pilot

6 weeks
from EUR 16,000 / USD 18,000
  • Working retrieval system on your own hardware
  • Citation verification on every answer
  • Measured evaluation bank — not a demo
Book a 30-minute call

Ongoing engineering

Rate
EUR 110 / USD 120 per hour
  • Rust system design and implementation
  • Inference infrastructure and cluster operations
  • For work outside a fixed package
Book a 30-minute call

Case studies

Case studies

Legal RAG with citation verification

A production legal assistant that answers from a corpus and verifies every citation before returning it, scored against a measured gold-question bank (38/50 at last measurement, with a written plan to 45/50). The evaluation bank is the deliverable that makes quality arguable instead of anecdotal.

A 4-node GPU platform for USD 65–80/month

Sovereign LLM inference on 2× RTX 3090 plus a cloud edge node, with automatic database failover and cross-node service placement. On-premise is not more expensive than cloud — for steady inference load it is dramatically cheaper.

From zero to production in days

Two public healthcare/services applications (Colombialifecare, Esperanza) taken from empty repository to live production, in Rust, server-rendered, with no JavaScript framework. Sovereign does not mean slow to build.

The stack in production

100% Rust production stack Local LLM inference on own GPUs Multi-node cluster with automatic failover 10 live MCP servers Server-rendered, zero JS frameworks

Book a 30-minute call

Book a call