ML engineer, researcher, independent thinker.

I'm a machine learning engineer and research scientist working at the boundary between structured symbolic reasoning and learned neural representations. Currently at The Network Group, I build knowledge extraction pipelines for high-stakes domains — legal, financial, and operational environments where reliability and explainability are requirements, not nice-to-haves.

My work at The Network Group has focused on deploying ML systems in contexts where the cost of error is high: a false positive in a safety-monitoring system means wasted resources and eroded trust; a false negative can mean missed obligations or overlooked risk. This has shaped my research interests toward hybrid approaches — systems that combine the pattern-recognition capabilities of modern neural networks with the verifiability and compositionality of classical symbolic methods.

My academic background is in linguistics and computer science, and I am currently preparing applications for PhD programmes in neuro-symbolic AI and knowledge representation. I am particularly interested in programmes with strong connections to both the formal and applied sides of the field — where the theoretical questions (what is a representation? what does grounding require?) are taken seriously alongside the engineering challenges.

---

Experience

---

Intellectual interests

Alongside the technical work, I read and write about continental philosophy — particularly phenomenology and critical theory. I find the questions these traditions raise about mind, meaning, and recognition genuinely relevant to the problems I work on technically, even when the connection is oblique. Fanon on recognition, Heidegger on being-in-the-world, Wittgenstein on rule-following: these are not metaphors for ML problems, but they illuminate what the problems are about in a way that purely technical literature often doesn't.

On the more speculative end of ML theory, I'm interested in the dynamics of training — particularly in what Hurst exponents and long-range dependence in training loss curves might tell us about the structure of learning. This is very much at the "interesting question" stage rather than the "publishable result" stage, but I return to it.

---

Now →

---