Skip to content
Aleksandra Karnoz homeLinkedIn

Aleksandra Karnoz

Marketing and Technology Leader, exploring where creativity, curiosity, and AI meet.

Illustration of Aleksandra waving hello

Marketing work, shaped by curiosity.

Illustration of Aleksandra working on a laptop

Aleksandra works across marketing strategy, brand, and events, with a growing focus on where technology and digital tools fit into that mix, at Expert Technologies Group.

Brand
Shaping how the company shows up, from visuals to voice.
Events
Building relationships and representation at industry events.
Digital
Managing the websites, campaigns, and channels that carry it all.

  1. 2022–2025

    BSc Psychology

    University of Warwick. Final mark: 2:1.

  2. Sep 2025

    Marketing Intern

    Joined Expert Technologies Group in Coventry.

  3. Jul 2026 to present

    Marketing and Technology Leader

    Progressed from the internship into her current role.

Working with AI, Practically.

Where this started

It began with gaming, or more specifically, the hardware behind it: what made a machine actually run well, and why. That curiosity grew outward over time, into how technology companies compete and build advantage, and eventually into academic work. Her dissertation examined how trust is built, and sometimes manipulated, on social media, work that ended up anticipating regulatory shifts that followed in that exact space. Around the same time, a business analysis of NVIDIA's AI-driven gaming technology looked at how a company builds real advantage around a fast-moving AI capability. The same question has followed her since: not just how to use AI, but how to use it meaningfully.

Practical work

Aleksandra is genuinely curious about how far AI can push marketing itself, not just content creation, but automation, reporting, and the effectiveness of strategy more broadly. That's pushed her beyond chat-based AI tools and into IDE and CLI environments, refining real websites and tools through testing and iteration rather than coding from scratch.

Email signature generator

Built with Claude Code, iterated with feedback from directors and engineers, and rolled out as part of onboarding.

Freelancer landing page

Built and deployed a landing and portfolio page for a freelancer using Claude Code, Vercel, and Supabase.

Lead organisation tool

Built a tool to manage post-event lead allocation, using Claude Code.

This site

This site didn't start as a portfolio. It began as a different idea entirely, and somewhere in the planning process, it became clear that what she actually wanted to build first was simpler: a website about herself, built properly with AI coding tools, not prompted into existence in an afternoon.

Planning first

Before writing any code, the project started with a set of markdown documentation files, an AGENTS.md file acting as the anchor, laying out what the site was for, how it should look, what it should be built with, and how it should stay maintainable. Everything the coding agent built afterward had to trace back to that document.

Research, not guessing

Design decisions were grounded in actual research, not instinct. That meant studying how SaaS companies use UX and UI to communicate, Apple's restraint and use of imagery in particular, and just as importantly, naming the visual patterns to actively avoid: the cream-and-terracotta look, the neon-on-black look, and the small “eyebrow” label above a headline that instantly reads as AI-generated right now.

Design, on purpose

The result was a deliberately disciplined identity: black and white, no accent colour, with Inter for headlines and body text and Geist Mono for structural details like dates and labels.

Development workflow

Working properly also meant working like a developer would: branches instead of pushing straight to main, a proper GitHub workflow, and deployment through Vercel with instant rollback if anything broke. Security got real attention too: dependency hygiene, keeping secrets out of the codebase, and only installing AI agent skills from verified sources, after learning how many public ones turn out to be malicious.

Built to last

A running change log tracks everything built after launch, and new features start as a short written brief before any code gets touched, planning as an ongoing habit, not a one-off step.

What comes next

This is the first project built this way, and it won't be the last. A contact form and some backend work are next, most likely using Supabase, along with deeper context mapping, alongside ongoing coursework to keep pushing further into how AI tools actually get used, not just talked about.

Next.jsReactTypeScriptTailwind CSSVercelMotionGSAP

My AI Roadmap

Matching the tool to the task

Not every task needs the most powerful model available. Day-to-day coding runs through Codex inside Cursor, a capable, efficient setup for building. Reviewing complex features brings in stronger models, Grok and Claude Opus, and full code audits use the most capable model available, Claude Fable. The most powerful tools only get used when the task actually calls for it.

Context engineering

Now working with context engineering: structuring markdown documentation so an AI agent understands a project properly before building anything.

Chatbot AI
Coding environments
Context engineering
MCP connections
Structured data
Instructions / context
Neo4j graph
Chained tasks
  1. 1

    Chatbot AI and prompt engineering

    Started with chatbot-based AI and prompt engineering, learning how to ask well and structure requests to get useful output.

  2. 2

    Coding environments

    Moved into coding environments, working through an IDE rather than a chat window.

  3. 3

    Context engineering

    Now working with context engineering: structuring markdown documentation so an AI agent understands a project properly before building anything.

  4. 4

    MCP connections

    Exploring MCP (Model Context Protocol) to connect projects directly to other tools, like Figma, enabling more automation and a higher standard of output across the workflow.

  5. 5

    Structured data

    Testing how structured data changes an AI agent's answers, comparing plain documentation against structured formats like YAML.

  6. 6

    Instructions and context

    Separating instructions from context: keeping "how the agent should behave" clearly apart from "what it needs to know."

  7. 7

    Neo4j knowledge graph

    Building a knowledge graph in Neo4j, mapping how a project's information actually connects, and having an agent query it directly rather than reading full documents.

  8. 8

    Multi-step tasks

    Chaining that together into multi-step tasks: querying, summarising, and drafting in sequence.

Thoughtful about the work around the work.

Ownership with context

Takes real ownership of projects, planning thoroughly and adapting as priorities shift.

Making complexity clear

Brings an outsider eye to engineering content, helping keep specialist material understandable for non-specialist audiences.

Growing through feedback

Takes feedback well, and treats it as part of getting to a better outcome rather than a personal steer to defend against.

Curiosity has a long history.

Illustration of Aleksandra holding a book and camera

Trend literacy

Built a TikTok account to 1 to 2 million+ likes, with multiple videos passing 3 million views.

Photography

Travels and shoots regularly. At Warwick Photography Society, served as Vice President and Social Media Manager, and taught members Lightroom.

Technology and business history

A sustained interest that started with gaming hardware and has grown into following the semiconductor and AI infrastructure space more broadly, through books like Chip War and Power Play.

Society marketing

As Head of Marketing for Warwick Russian-Speaking Society, ran content strategy, analytics, event promotion, and engagement workshops.