Stoycho Lyubenov

Hey there, I'm

Stoycho Lyubenov

Building personalised AI workflows, automations and agents around how your business actually runs.

Available for new projects

Book a Discovery Call 10 yrs sales · 5 yrs AI · London
Stoycho Lyubenov

I turn manual, repetitive work into AI systems that run themselves — built for your business, not from a template.

0+

Years in sales

0+

Years in AI

0+

Years building websites

0

Stage delivery process

24/7

Managed support

I'm Stoycho, a 33-year-old AI automation consultant based in London. I spent ten years in sales before moving into the AI sector, where for the last five years I've been designing and shipping workflows, automations and agents of every shape.

That sales background matters: I talk in terms of revenue, hours saved and ROI — not model benchmarks. I'm obsessive about detail, and every workflow and agent I build is personalised to how your business actually operates.

I also design and build websites from scratch — sitemap and wireframes, the words and the order they're read in, the colour and type system, the motion, and the performance, SEO and accessibility work that makes it production-ready. This site is one of them.

How I work

  • Claude, n8n and Google Workspace as the core stack
  • Working prototypes in a sandbox — not slide decks
  • Every integration point mapped before code is written
  • Ongoing optimisation once you're live
See the process

Core expertise

02

Workflow Automation

End-to-end business processes automated with n8n and your Google services — triggered, routed and reported without a human in the loop.

  • n8n pipelines
  • Google Workspace
  • API integrations
  • Scheduling & triage
03

AI Agents & Orchestration

Agents that call tools, remember context and coordinate with each other — designed with proven agentic patterns.

  • Function calling
  • Multi-agent systems
  • LangGraph
  • MCP
  • Agent memory
04

LLM Apps & RAG

Chat and search over your own documents with retrieval that actually finds the right answer.

  • Prompt engineering
  • Embeddings
  • Vector databases
  • Document chunking
  • Advanced retrieval
05

LLMOps & Evaluation

Deployed, monitored and measured — so you know your AI is working, not just running.

  • Deployment & CI/CD
  • Observability
  • LLM-as-a-judge
  • RAG & agent evals
  • Benchmarking
06

ML Foundations

Classic machine learning where it beats an LLM — with proper pipelines, tracking and production monitoring.

  • scikit-learn
  • Data quality
  • MLflow
  • DVC
  • Production monitoring
07

Safety, Guardrails & Governance

AI you can trust in front of customers — with security, bias checks and governance frameworks built in.

  • Responsible AI
  • LLM security
  • Guardrails
  • Data ethics
  • Governance

Every project runs through four stages

  1. Stage 1

    Discovery Audit

    We deep-dive into your manual workflows to find high-ROI automation opportunities — forensic clarity on where AI delivers maximum leverage.

  2. Stage 2

    Architecture Mapping

    We design a custom blueprint connecting Claude, n8n and your Google services — mapping every integration point before a line of code is written.

  3. Stage 3

    Rapid Prototype & Build

    A functional system is deployed in a sandbox for testing. You see working systems fast — not slide decks and promises.

  4. Stage 4

    Launch & Managed Support

    We integrate into your live environment and provide ongoing optimisation. Your automations run 24/7 — and so does our commitment.

From blank page to production

When the project is a website, this is what happens inside those four stages — seven steps, in this order, every time.

  1. Step 01

    Discovery & Sitemap

    Who the site is for, what it has to make them do, and every page and section needed to get there. The information architecture is agreed before anything is drawn.

  2. Step 02

    Wireframes

    Layout and hierarchy in greyscale — what sits where, what comes first, what each screen weighs. Structure is settled while it's still cheap to change.

  3. Step 03

    Copy & Information Style

    What each section actually says, in what order, in what voice. Every line is checked for clarity and scannability, because a beautiful page that doesn't explain itself has failed.

  4. Step 04

    Colour & Visual System

    A palette with real contrast ratios, a type scale, spacing tokens and reusable components — a system, not a set of one-off screens.

  5. Step 05

    Motion & Interaction

    Scroll reveals, micro-interactions and transitions that direct attention rather than perform. Everything degrades cleanly for reduced-motion and low-power devices.

  6. Step 06

    Optimisation

    Core Web Vitals, compressed assets, responsive behaviour on real devices, SEO metadata, and accessibility — contrast, focus states, semantics and keyboard paths.

  7. Step 07

    Launch & Production

    Built, deployed, analytics wired up and handed over with documentation — then iterated on with real numbers instead of opinions.

Everything I work with

01 LLMs & RAG 8 areas
  • LLM fundamentals
  • Prompt engineering for devs
  • Working with LLM APIs
  • Embeddings & semantic search
  • Vector databases
  • Retrieval augmented generation
  • Document processing & chunking
  • Advanced retrieval
02 Integration, Agents & Orchestration 10 areas
  • Function calling & tool use
  • Structured outputs & validation
  • LLM apps with frameworks
  • Agent fundamentals
  • Agentic design patterns
  • Orchestration with LangGraph
  • Multi-agent systems
  • Agent memory & state
  • Agentic RAG
  • Model Context Protocol (MCP)
03 Ops & Evaluation 10 areas
  • LLMOps fundamentals
  • Serving open source models
  • Deploying LLM applications
  • Observability & monitoring
  • CI/CD for LLM applications
  • Evaluation fundamentals
  • LLM-as-a-judge
  • RAG evaluation
  • Evaluating AI agents
  • Benchmarking models
04 ML Foundations 7 areas
  • ML fundamentals & metrics
  • scikit-learn pipelines & training
  • Data engineering & quality
  • Experiment tracking with MLflow
  • Data versioning with DVC
  • Deployment & CI/CD
  • Monitoring ML in production
05 Safety & Ethics 5 areas
  • Responsible AI practices
  • Data ethics & bias
  • LLM security & risk
  • Guardrails
  • AI governance & frameworks
06 Web Design & Front-end 9 areas
  • Wireframing & prototyping
  • Information architecture
  • UX copywriting & tone
  • Colour & type systems
  • Design tokens & components
  • Motion & micro-interactions
  • Performance & Core Web Vitals
  • Accessibility (WCAG)
  • SEO & metadata

Let's automate the work that's slowing you down.

Start a project
  • ROI-first
  • Sales-trained communicator
  • Working prototypes, not decks
  • Claude + n8n + Google native
  • 24/7 managed support
  • Detail-obsessed
  • Bespoke, never templated
  • Runs in your live environment

Tell me what's slowing you down