Powering Your Business With AIFrom Idea to Implementation

Working across the full content pipeline — prompt engineering for pharmaceutical compliance, multilingual generation, RAG over approved medical documents, and AI image generation with brand-compliant styles. Building systems where the AI understands not just what to write, but what it is and isn't allowed to say in a heavily regulated industry.

The hard part was not calculating emissions — it was the matching. Invoice line items are messy, inconsistent, and written by hundreds of different vendors. Climate databases like Klimadatabasen describe products in precise scientific terms. Getting those two worlds to meet reliably, at scale, across all product categories, is where the real engineering challenge lies. Driven in part by EU regulation that simultaneously makes CO₂ reporting mandatory and technically complex.

Three compounding problems: natural class imbalance — far more people wish for makeup than specialist hardware, so perfect balance is neither achievable nor desirable; products that genuinely fall under multiple taxonomy branches simultaneously; and recognising that the same product appearing on different retailer websites is actually one item, despite completely different titles, images, and pricing. Solved using a combination of ML models and LLMs.

Two compounding challenges made this technically hard. First: the product database spans thousands of product types, each with different attributes, features, and compliance rules — the model had to understand what makes a valid match for a kettle versus a laptop versus a sofa. Second: as a data aggregator, the client ingests product data from sources all over the world — meaning the incoming data was massively inconsistent in structure, format, and completeness. The system had to work reliably across all of it.

Rather than analysing what predators say, the thesis examined how they behave — using graph structure to surface anomalous social network patterns that linguistic models alone would miss. Anomalous users exhibited statistically distinct behaviour in terms of their network connections and interaction patterns.

Forecasting drivers included holidays, day of week, product promotions, store closures, and global geopolitical events — built to be robust in conditions where last week's data was no longer a reliable signal for next week's reality. The model had to reason about what it didn't know as much as what it did.

Working across Adobe Analytics and Google Analytics for different companies simultaneously builds a rare cross-platform fluency — you stop thinking in tools and start thinking in questions. What does this business need to know, and what is the cleanest path from raw data to that answer?

A research project investigating whether a model could be trained from product catalogue images alone to understand the visual nuances of jewellery — distinguishing hoop earrings from bangles, rings from pendants. No off-the-shelf AI APIs, no foundation models. A neural network built and trained from the ground up to see jewellery the way a customer does.

"Thinking agentically is not just for AI engineers — it's a way of structuring our work so we are ready for the future. The better we understand our own processes, the easier it becomes to hand them over to AI agents." — Natasha Norsker, AI Agentbogen

"If you think of AI projects as venture investments, then it's not a bad number that two thirds don't go further. It only becomes a problem if you don't learn enough from them, don't stop them fast enough, or don't define exit criteria upfront. AI work should be managed as a portfolio — many small bets, fast validation, brutal selection, clear scaling of winners." — Natasha Norsker

"Computers have become all-pervasive — and that has pushed a million Danes over the edge, leaving them unable to participate in society. No one asked for digitalisation, but it has been forced upon us to such a degree that you cannot function in Denmark without the latest smartphone and computer." — Radio 4, Tager jeg fejl

"AI doesn't give you talent. But I do believe it can help you stay in your talent zone — carving out time for deep work, helping you stay curious, and letting you see ideas from different perspectives." — Natasha Norsker

"Natasha Norsker shares how more women are landing within AI — and how there seems to be a new focus on sharpening interest in tech from girlhood onwards." — PROSAbladet

"Together they explored how AI can elevate the success of startups and the role women play — covering mentorship, bias in AI systems, and how to take your place in a field that's moving fast." — Connected Women in AI

"Get insight into how companies can transform data into valuable business insights — learning from real experience and finding new ways to leverage data." — Computerworld Cloud Festival

"One of the best things about specialising in AI is that you can apply the same skills across completely different domains. The same algorithms that work well for lawyers can also be applied in waste sorting, agriculture, and much more — and discovering how use cases transfer from domain to domain is one of the most exciting parts of the work." — Natasha Norsker

"We are live at Applied AI Nordics 2024 at DR Koncerthuset — Natasha Norsker shares her framework for getting started with AI: technology, safety, and the user. Not a new framework — but one that's consistently overlooked when organisations chase the technology before understanding the problem." — Verbos Podcast

"Thank you to Natasha Norsker for her in-depth session 'Let's Get Practical on AI' — the perfect ending to a content-rich day, full of learning and new perspectives." — Applied AI Nordics

"Using Gartner's model as a foundation, they guided Bestyrelseskvinder's members through three decisive ways to create value with AI: Deploy — changes that ease everyday workloads; Reshape — improving heavy, time-consuming processes; Invent — new business models created with and by AI." — Bestyrelseskvinder

"Natasha er Co-founder af Women in AI Denmark og Data Scientist i Todai. Vi taler om AI før og efter ChatGPT, arbejdet som AI konsulent, graf neurale netværk og meget mere." — Verbos Podcast

"Dedicated to empowering women and fostering open conversations about equality, personal development, and leadership — addressing gender disparities in policy-making and the technology industry across Europe." — Microsoft Conversations Unscripted

"From ChatGPT to DALL-E, AI and machine learning are evolving at a rapid pace — impacting data centers both in operations and in the ever-growing IT loads that AI applications require and generate." — Datacenter Forum Copenhagen

"Awarded to an individual who has made a meaningful and lasting contribution to software development — achieving outstanding projects and developing creative solutions to problems solved with code." — Nordic Women in Tech Awards

Nordic Women in Tech Awards has celebrated over 2,600 female role models across 11,000+ members since 2019 — run entirely by volunteers and one of the fastest growing changemaker communities in the Nordics.

"As Founder of Women in AI Denmark and AI Consultant at Todai, Natasha is breaking down barriers and paving the way for women in tech. From developing groundbreaking AI solutions to creating a community of female experts, her influence is shaping the future of AI in Denmark. Her mission? To elevate women's voices in the field — and she's doing just that." — Nordic Women in Tech Awards

Founded Women in AI Denmark to create a space where women working with AI could meet, connect, learn, and grow together — ensuring they are seen, heard, and supported in a field where they are still underrepresented. Alongside voluntary work on DTU's AI and mathematics study council and serving as a volunteer mentor for students pursuing careers in AI.