I build the tools other devs build with. Authoring systems, visual editors, no-code platforms, and the compilers and toolchains underneath them. Fifteen years across Flash, Java, C#, TypeScript and Python, whichever the tool needed.
Agents got good enough at implementation that the work can finally move up to spec and architecture. That only holds while the tests can be treated as ground truth, so most of what is below is about validating tests.
alextavor.github.io · LinkedIn · Utrecht, NL
pdd-experiment-cave: a three-arm experiment comparing a raw agent against a spec-driven method on a large brownfield codebase. Prompts frozen, commits pinned, blind adjudication by a separate judge. Stopped after run 1, because the result was stark and a run is expensive: the method found the same four high-severity issues as the raw agent at 3.9x the cost. Dashboard.
proof-driven-development: concluded. A method and CLI for changing code you cannot verify by reading, legacy or AI-generated, without breaking it. The experiment above is why it stopped, and the post-mortem says what came out of it.
engineering-discipline: the rules that survived, as an installable, stack-independent plugin. Every one exists because something went wrong on a real project, and each cites the commits it was recovered from.
chdr: a LangGraph orchestrator for an autonomous TDD loop, with a post-mortem on the results.
cave-public: a browser simulation game on a custom data-driven engine, with compiler and linker, physics, simulation and narrative subsystems, an authoring suite, and the art tool I made its visuals with. Play it.
dod: one local supervisor that registers, launches and frames every other local dashboard in a single admin UI.
rapport: local-only analysis and graphing of relationships from text chat dumps.
Stack: TypeScript · React · Node · Python · C# · Java · Unity
Open to lead and senior engineering roles. Reachable on LinkedIn.