Rafael Lopes

Founder & Principal AI Engineer

Vancouver, British Columbia, Canada

Founder of Exaflop — sovereign, local-first production AI, built and run end to end on hardware I own: a 3-node Kubernetes cluster, a GitOps pipeline, ~184 GB of distributed LLM inference, and a hybrid-RAG stack gated to 100/100. The running system is the proof, not the claim.

Surfaces

• BlogPosts + twice-weekly research briefs, RAG-authored and citation-gated.
• ExaflopSovereign research copilot over ~38,000 Canadian public-research documents.
• InfraThe 3-node K3s cluster + GitOps pipeline, documented.
• GitHubSource.
• LinkedInProfile + experience.

Selected work

• Distributed K3s cluster + GitOps operation3-node K3s cluster (x86 + ARM), GPU scheduling, Argo CD GitOps, Terraform-managed Cloudflare Zero Trust — built and operated solo.3 nodes184 GB inference2025-04
• Hybrid RAG engineeringSovereign research copilot over ~38,000 documents: BM25 + dense + RRF + cross-encoder rerank + citation verification, held to a deterministic 100/100 quality gate.38,000 docsgate 100/1002025-06
• AI cost governancePer-agent + per-model token metering, a self-validating savings ledger, policy gates and caching — ~85% LLM spend avoided without degrading answers.85% cost reductionquality impact: none2025-09
• AI-native web architectureZero-client-JS site with perfect Core Web Vitals, full JSON-LD + llms.txt + an isolated semantic-search service — the website as a machine API.0 KB client JSCWV 100/1002025-10

Offers

• AI system design
• RAG pipeline architecture & optimization
• LLM cost reduction
• Agent-readiness (CAV) audits
• Platform engineering advisory
• Web performance engineering

Expertise

• Production AI
• Retrieval-Augmented Generation
• Distributed LLM inference
• AI efficiency
• AI cost governance
• Web performance
• Core Web Vitals
• Kubernetes
• Argo CD
• GitOps
• Platform engineering
• Site Reliability Engineering
• Observability
• Terraform
• Design systems