Optimization is a research problem. We treat it like one.

Finding the best version of a real system, across competing objectives, on real hardware, is a hard search problem. TurinTech's research is where our methods come from: search strategies, evaluation, and validation for optimizing production systems. What works here gets built into Artemis.

Eight years of teaching software to improve itself.

TurinTech began as a UCL spinout researching how software could continuously improve through evolutionary optimization, years before today's wave of generative AI.

The line of work is unbroken. The search methods we published in 2018 are the direct foundations of the ones running inside Artemis today, now combined with LLMs and applied to code, models, agents and infrastructure.

2018
Evolutionary search proves out
Darwinian Data Structure Selection (FSE) shows genetic search delivering measurable time, CPU and memory wins on real Java projects.
2020–21
Search gets smarter and broader
IEO brings ML-guided evolutionary optimization; genetic optimization extends the approach to production C++ libraries.
2022
From papers to platform
The Artemis ML yellow paper describes a full evolutionary AI studio: automated data processing, model optimization and multi-objective code optimization.
2025
The LLM era
Multi-LLM optimization, meta-prompting and automated agent tuning: the methods now built into Artemis.

Publications & technical reports.

Our published work on proprietary algorithms, evolutionary AI, proven results, and patented technology: genetic improvement and search-based software engineering, published years before LLMs existed and still compounding.

Conference