Cristian Dinu
  • Home
  • Writing
  • Talks
  • About

Cristian Dinu

Research software, energy, AI and society. Writing and work by Cristian Dinu, a Senior Research Software Engineer at UCL.

Research software · Energy · AI · Society

Engineering.
Uncertainty.
And being human.

I’m a software engineer working across energy, data and research — currently building the infrastructure behind UK household energy studies at UCL, and testing how AI systems and agents behave under pressure.

A little more about me

Cristian Dinu
Engineer, researcher,
occasionally wrong.

The public notebook

Recent writing

All writing

What has been keeping me busy, September 2026

AI is at the centre of what I’ve been doing, unsurprisingly. But testing and evaluating LLMs is only half the story — the other half is why I still insist on writing every word of this myself.

25 Sep 2026

From Bitcoin to AI: 9 of My Predictions That Aged Badly (and a Few That Didn’t)

I’ve made tech predictions that soared and others that crashed: from laughing at Bitcoin to betting on Linux. Turns out I’m sometimes good at spotting solid tech, bad at predicting people. My next bet? Generative AI + Model Context Protocol will redefine software.

29 Jul 2025

Dashed — Writing Like a Machine

I like dashes of all kinds: the hyphen, the syllable-breaking dash, the dialogue dash. I like diacritics, I like commas. Whenever I can, I use non-breaking spaces between words that belong together. I’ve been using them for decades, on my own, as consistently as I can.
23 Jul 2025

Power Shift: From Passive Consumers to Grid Players

Energy flexibility means the ability to adjust power generation and demand, so that they match exactly. It’s going to be a big topic in the following years, and it is very likely to impact our comfort, our wallets, but also our climate.

22 Jul 2025

AI: Our Chilling and Thrilling Child

Insights from Stephen Fry and Yuval Noah Harari at the Octopus Energy Tech Summit on AI as humanity’s child

28 Jun 2025
No matching items

Open questions

Thinking through, not finished with.

Lately I keep coming back to the same question: how do you evaluate and test software built on non-deterministic components, particularly LLMs and agents?

  • How do you establish what “good enough” means?
  • How do you investigate failures, and tell whether a change has actually improved the system?
  • What can simulations tell you about the real world?

More about my current interests

Ideas in practice

Work & talks

UCL · Energy research

Data systems for SERL & EDOL

My work focuses on designing and building data infrastructure that gives UK researchers access to detailed energy datasets while maintaining strong privacy protections.

About the work · EDOL update

RSECon25 · Talk & demonstration

The End of Front-End

Delivering software via LLMs and the Model Context Protocol.

Talk & materials · Demo code

© 2026 Cristian Dinu
Views here are my own.

 
  • RSS

  • Stochastic Hippo