holsee
I'm 20+ years deep building and shipping software of all shapes and sizes, commercial and open source.
I have a particular interest in distributed, fault-tolerant, concurrent systems and applied AI. My go-to weapons of choice centre around the BEAM for this very reason (if you know you know), but I gladly reach for the correct tool for the job. At one point in my career I left big finance and took a pay cut to join a cyber-security firm for no other reason than to get professional exposure as an Erlang engineer. I enjoy programming languages, language design and theory, and I don't feel constrained to individual specialisations, although for over a decade I went extremely deep in the .NET space, and for another decade have been shipping solutions oriented around the BEAM and Elixir as a default. It has been a good time.
I'm riding the agentic wave as an accelerationist and an optimist, with a healthy dose of experience in establishing AI ethics boards and an applied understanding of alignment. I enjoy pushing the limits of what AI can do in engineering and product, having spent years deep in research seeking generational improvements on early models through the dark arts of platform-level prompt engineering and hackery: less important now with the state of the art, but useful nonetheless when it comes to getting the most from smaller models.
I believe successful systems come from good, low-friction methods of collaboration between contributors, including in commercial settings. The role of engineering is changing rapidly. In a world where the execution of code authorship can be accelerated through agentic intelligence and process to a point where it is mechanical: I'm a firm believer in the importance of talented engineers and the value they bring to the table with their experience, problem solving, creativity, taste, inspiration, rigour, design/critical thinking and knack for elegance to produce the pillars that hold up systems which in turn hold up organisations.
This site covers a good chunk of my experience and the projects that I'm able to share in the open. There is more about me on the about page.
Start here
All writing →- conjure: Agent Skills for Elixir: Anthropic's Agent Skills spec in Elixir, with one Session API over four execution backends.
- flywheel: high-performance data channels: a bounded, back-pressuring channel on
:atomics, and why an unbounded mailbox fails late and all at once. - Control planes in agent SDLCs: I deleted the CAS store, the leases and the reconciler, and shipped on a GitHub substrate instead.
- Radix sort on sixteen cores: why a radix pass parallelises and a comparison sort doesn't, in sixteen .NET threads and sixteen BEAM processes.
- Growing a test language: Guy Steele's talk in words of one syllable, and the language design theory I borrowed for a test framework grammar.
- deep_eval_ex: LLM-as-judge metrics: my port of Confident AI's DeepEval to the BEAM — seven LLM-as-judge metrics running as native, concurrent evals inside
mix test. - exile: a database that grows its own schema: the URL is the schema and collections exist because you wrote to them — 48 hours on ETS and Channels, and it won Spawnfest.
The long view
- Timeline: every position, project, talk and library since 2007, dated, from the same files as the CV.
- Now: what is on the bench this month.