Hi! đź‘‹
I’m Natalia Elvira Astoreca, and this is where I share what I’ve learned about building AI systems that actually work with complex text.
What You’ll Find Here
This blog is about the practical side of AI engineering - the decisions, trade-offs, and linguistic complexity that determine whether an AI system succeeds or fails in production.
I write about:
- Data curation & annotation - Building datasets that capture real-world complexity
- Architecture decisions - When to use LLMs, fine-tuned models, or traditional NLP
- Linguistic patterns - The edge cases and structures that make or break text AI
- Production challenges - What actually happens when you deploy these systems
Who This Is For
If you’re working on AI systems that deal with text - whether you’re building them, evaluating them, or trying to understand why they behave the way they do - you might find something useful here.
The posts assume some technical background, but I try to explain concepts clearly. If something isn’t clear, that’s on me, not you.
A Bit About My Background
I come from an unusual path: I started with Ancient Greek linguistics, studying archaic alphabets and dialect variations at Cambridge. That taught me to see language as a system full of patterns, exceptions, and context-dependent meaning.
These days, I work on modern text AI systems, but the core challenge is remarkably similar - understanding how language actually works, not how we think it should work.
If you want to know more about my work or background, check out the About me page.