Manuel Porras

Manuel Porras

AI Agent Engineer · Claude Code, Multi-Agent Systems and AI Automation

Berlin · remote · [email protected]
manuelporras.com · github.com/porrasbrand · linkedin.com/in/manuel-porras

Profile

I build and run fleets of AI agents in production: currently 38 Claude Code agents on two servers, including a dedicated expert agent for each of 18 agency clients. Work is routed through message queues, agents are monitored and recover automatically after failures. Before that, 20 years of digital marketing and web development for US businesses, focused on generative AI since late 2022. I know which processes are worth handing to agents, and I build them to run unattended.

Core Skills

  • AI agents: Claude Code, custom multi-agent orchestration (Node.js), one agent per client, remote control
  • Agent operations: Linux server operations, message queues, monitoring, automatic recovery
  • LLMs: OpenAI, Anthropic Claude, Google Gemini · prompt engineering · structured outputs
  • Automation: n8n (self-hosted), LangChain agents, API integrations, RAG (Pinecone, pgvector)
  • Private AI: local LLMs (ollama), speech recognition with faster-whisper, GDPR-compliant

Professional Experience

Breakthrough3x, USA — AI Automation & Process Optimization

03/2020 – present · remote · freelance
US marketing agency for home-services businesses (HVAC)

Since 01/2024: Head of AI Automation

  • AI agent fleet: 38 Claude Code agents on two servers running 24/7, including an expert agent for each of the 18 agency clients; work arrives via Slack and message queues at the right agent
  • Self-healing operations: checks every minute; crashed agents are rebuilt automatically, with protection against restart loops; new client agents set up by script in minutes
  • AI meeting analysis: Zoom calls turn automatically into tasks and summaries in Slack and ProofHub — 30+ client meetings per week, results in 15 minutes instead of up to 2 days
  • Private call analysis (2026): 266 customer calls of an HVAC business transcribed and analysed without any data leaving the premises; refuted an incorrect client assumption with data
  • PPC monitoring console: budgets, cost per lead, lead sources and tracking for the whole client portfolio, checked every morning; read-only; in an audit the real cost per lead was about 60 % above the reported one
  • Dashboards that consolidate ad, meeting and project data automatically; lead hand-off from Meta to ServiceTitan with no manual step

03/2020 – 12/2023: SEO, PPC & Automation

  • SEO, Google Ads and local search for US HVAC businesses; automated reporting

Awesome Medical, Las Vegas (USA) — Partner, Digital Marketing & Web Development

05/2006 – present · remote · freelance, with a US business partner
Aesthetic surgery clinics in several US cities

  • RAG chatbot platform (2025): self-hosted n8n crawls the clinic websites, stores the content in Pinecone and answers patient questions on prices and treatments, with links and before-and-after photos. My own crawler replaced the paid service.
  • AI content engine (2023): automated blog articles for client websites; solved the output limit of the models of the time by generating paragraph by paragraph
  • Since late 2022: generative AI for campaign analysis, ad copy and landing pages
  • Websites, SEO, Google Ads and Facebook campaigns for clinics with combined revenue of over USD 1.5M per month (2018); #1 on Google for the most important search terms

Further AI Projects

  • AI pipeline on own GPU (2026): speech recognition, text-to-speech and agents for audio production, research and SEO on a second server, operated via remote control
  • White-label AI platform (2025): 30+ AI assistants in one environment; new clients by configuration instead of code
  • Quality measurement for AI output (2026): blind tests show whether a change really improves results — 56 documented test cycles
  • 3D scenes written as code (2026): agents build explorable 3D sets from open geodata (OpenStreetMap, Berlin LoD2) or two photos, as Blender code — 30+ sets in 10 days

Details with architecture: manuelporras.com/work

Technologies

JavaScript/Node.js, Python, TypeScript, PHP, SQL · PostgreSQL, MySQL, SQLite · Linux, tmux, pm2, cron, Docker, GPU/CUDA, Apache, nginx, Cloudflare · Playwright, WordPress, REST APIs

Education

M.A. Technology and Innovation Management · 2007 – 2010
FH Brandenburg and Brandenburg University of Technology Cottbus, Germany

Degree in Systems Engineering (Diplom, 5-year program) · 1994 – 2002
Universidad del Valle, Cali, Colombia

International Baccalaureate · 1991 – 1993
Deutsche Schule Cali, Colombia

Languages

Spanish (native) · German (C2) · English (professional) · Italian (fluent) · French and Polish (basic)

Work authorization: German EU long-term residence permit with unrestricted right to work · citizenship application pending