Evomedia.net

The natural evolution of your business.

Expertise

Core skills — Kelly Michels: LLM & RAG platform engineering, multi-tenant SaaS, AI-assisted delivery and testing, software design and UI design, plus AWS operations.

LLM & RAG engineering

Designed, built, and operate evo-ai, a multi-tenant RAG platform running in production as the Ask AI feature inside evo.ehs: retrieval with citations grounded only in an organization’s own records, text-to-SQL analytics over live schemas, guardrails, and a retrieval evaluation harness (golden sets, mean reciprocal rank) so tuning never silently regresses quality. Qdrant vector store, embedding pipelines, local and hosted model serving.

Multi-tenant SaaS & platform

Production SaaS with row-level tenant isolation on PostgreSQL, WebAuthn passkey sign-in, and Stripe subscription billing. Built evo.platform, a shared NestJS foundation — common authentication, per-tenant data isolation, and an SDK consumed by multiple apps — proving out under evo.provensheet and evo.docketmail.

AI-assisted delivery

Building and shipping production software with AI agents daily — Claude Code against a sixteen-repo fleet: token efficiency, prompt optimization, planning workflows, and model selection tuned for result quality against cost. Python/NiceGUI, Node.js/Next.js, and most common enterprise IDEs and APIs.

Software design & UI design

Product and system thinking, information architecture, interaction patterns, visual hierarchy, and cohesive UI for web, mobile, desktop experiences — from early structure through implementation-ready layouts, muscle-memory workflows and more.

Full-stack development

End-to-end application delivery across two stacks: Python / NiceGUI (SQLAlchemy 2.x, Alembic, PostgreSQL, psycopg, Stripe billing, fpdf2 PDF generation, httpx/requests for external APIs) for data-driven apps, and Node.js / Next.js (TypeScript, tRPC, Prisma, React, Vite) for modern web and mobile — from schema design through Docker deploy.

QA & automation — SDET

Automated functional, regression, integration, and end-to-end testing; visual validation and accessibility testing. Fleet-wide test infrastructure: every suite runs nightly with each individual test recorded and published on a live results board, coverage measured honestly across every repo, and AI-assisted triage of failures. Includes systematic QA of LLM systems — text-to-SQL correctness, retrieval, and guardrails — with a documented defect loop: evo.qaforaisystems.

CI/CD & DevOps

GitHub Actions PR gates, branch-protected trunk workflows across the fleet, Jenkins, Artifactory, Jira webhooks & triggers, optimized deployments, and repeatable release tooling.

API & scripting

API testing, Python, PHP, PowerShell, Cursor automation.

3D & AI domain

Autodesk AI MotionMaker, ML Toolkit, 3ds Max, 3D generalist, Maya pipelines, character animation validation, geometry workflows.

Support systems & infrastructure

Self-hosted mail (Postfix, Dovecot, Rspamd, Roundcube), SMTP relay via Brevo, monitoring dashboards (Prometheus, Grafana, cAdvisor, Node Exporter), automated TLS renewal, and operational tooling that keeps production stacks healthy around the clock.

Professional skills

Executive communication, technical documentation, cross-team collaboration, stakeholder alignment. Agile methodology in sprint-based workflows and similar iterative delivery frameworks.

AWS — setup, configuration, maintenance

Strong practical experience running production-facing sites and services on Amazon Web Services, end-to-end:

  • EC2: instance sizing, AMIs, SSH access, security groups, elastic IPs / DNS, patching rhythm, cost-aware choices. Running Ubuntu 24.04 LTS with Docker and Docker Compose as the container runtime for all production workloads.
  • CloudFront CDN: distribution setup for www.evomedia.net, origin protocol policy configuration, AllViewer origin request policy to forward the correct Host header, debugging redirect loops caused by ALB HTTP→HTTPS interplay.
  • ALB (Application Load Balancer): listener rules, target group forwarding vs. redirect modes, wiring ALB origins to CloudFront without creating infinite redirect loops.
  • Route 53 DNS: full zone management — A records, CNAME, MX, SPF TXT, and DKIM TXT records. Configured authoritative DNS for evomedia.net including subdomains mail, webmail, mail-admin, and dashboard.
  • Multi-container Docker stacks: four independent Compose projects sharing a single external bridge network (web) — edge proxy, mail stack, monitoring stack, and website — each deployable and restartable independently.
  • nginx edge proxy: single-container ingress routing traffic to all backend services by server_name, HTTP→HTTPS redirects, WebSocket upgrade passthrough, correct reverse-proxy header forwarding (Host, X-Forwarded-Proto, X-Real-IP).
  • TLS & certificate management: Let’s Encrypt via Certbot in Docker (webroot ACME challenge), multi-SAN certificates covering all subdomains, automated daily renewal via cron with nginx hot-reload on success.
  • Mail server stack: docker-mailserver running Postfix (inbound SMTP), Dovecot (IMAP/TLS), Rspamd (spam filtering), and Fail2ban (intrusion prevention). Outbound relay via Brevo SMTP to work around AWS port-25 egress block. Roundcube webmail and a custom Flask mail-admin UI.
  • Monitoring & observability: Prometheus scraping six targets (Node Exporter, cAdvisor, nginx-exporter, postfix-exporter, itself, Grafana), 30-day metric retention, Grafana 13 dashboards for host metrics, container metrics, nginx, mail, and an evomedia overview. Email alerting via Grafana contact points for CPU, memory, disk, and mail-queue thresholds.
  • Security hardening: EC2 security groups scoped to required ports only, Fail2ban whitelisting internal Docker subnets, nginx CSRF-compliant proxy headers for Grafana 13, HSTS on all HTTPS vhosts.
  • Release & maintenance: repeatable deploy scripts (zdeploy.ps1, deploy_ec2.ps1) using SCP + SSH from Windows, smoke checks after deploy, 2 GB swap file for memory headroom, routine troubleshooting of container networking, iptables, and Docker DNS resolution.

Technical interests

3D printing & digital fabrication — model preparation, slicing optimization, geometry validation, printer calibration.