I’m Natalia, an AI Engineer from Madrid (Spain) who accidentally found the perfect intersection between ancient languages and modern AI.
The Journey from Dead Languages to AI
My career started in an unexpected place: studying Ancient Greek alphabets at the University of Cambridge. I spent years analyzing how archaic Greek writing systems evolved, comparing linguistic patterns across dialects, and reconstructing how people actually wrote and spoke thousands of years ago.
That work taught me something crucial: language is messy, full of edge cases, and constantly evolving. You need systematic analysis combined with deep pattern recognition to make sense of it.
When I discovered computational linguistics and AI, I realized this was the same challenge - just with modern text and machine learning models instead of ancient inscriptions. The skills transferred perfectly: spotting patterns, identifying edge cases, understanding why standard approaches fail on complex language.
What I Do Now
I work independently helping companies build reliable AI systems for complex text problems. Most teams struggle to bridge the gap from raw text data to production AI systems - they either over-engineer with expensive LLMs when simpler approaches would work, or they skip critical design steps and end up with systems that don’t solve the actual business problem.
I combine three perspectives that most specialists working alone can’t:
- Linguistics and philology - I understand what makes text complex and how to make taylored corpora for specific problems
- Product thinking - I translate business needs into technical specs and speak both languages fluently
- ML Engineering - I build architectures that actually work in production
This combination means I see problems pure engineers miss, define requirements pure linguists can’t, and deliver systems pure PMs can only scope.
I’ve designed and built successful AI applications for complex text extraction in fintech and healthtech - industries where getting text wrong has real consequences. Currently, I work with companies like The Newsroom, helping them navigate the gap between “we have text data” and “we have a working AI system.”
How I Can Help
I work with companies that need to build text AI systems that actually work in production. This typically involves:
- Architecture Decisions: Traditional NLP vs. fine-tuned models vs. LLMs - and when to use each. Sometimes the answer is a cutting-edge LLM. Often it’s not. The key is knowing the difference.
- End-to-End Pipeline Design: From vague requirements to production-ready systems. I help you avoid the common trap of jumping to a technical solution before understanding the actual problem.
- Evaluation Frameworks: Metrics that actually matter for your business, not just academic benchmarks that look good in papers.
- Data Strategy: Curation, annotation, and quality control for complex text. Getting this wrong is the fastest way to waste time and money.
- Cost Optimization: Performance without burning budget on unnecessary LLM calls. I help find the right balance so you’re not killing flies with bazookas.
My Approach
I believe most AI projects fail not because of technology, but because of misalignment between business needs and technical implementation.
My approach is practical:
- Understand the actual business problem (not the assumed technical solution)
- Analyze what makes the text difficult
- Choose the simplest architecture that will work
- Build evaluations that measure what matters
- Optimize for both performance and cost
Beyond Work
When I’m not thinking about text and AI, I’m usually:
- Playing videogames (current favorites: “Papers, Please”, “Outer Wilds”, and “Stardew Valley”)
- Crocheting or knitting to balance out all the screen time
- Spending time with family
Work With Me
If you’re struggling to bridge the gap from raw text data to production AI systems, or wondering whether you need LLMs or something simpler, let’s talk.