Gil Raitses

Developer tooling

A line of research tools I built for labs and field work taught the same instinct developer tooling needs: show what people can build and smooth the path to adoption.

Each tool below came from a real research need. Taken together they are a record of building things, getting other people to use them and automating the work behind them, which is the same job a developer experience role does for a platform.

Shipped research tools

These projects each solved a measurement or conversion problem in isolation, and together they show how a research toolchain turns raw capture into validated data.

  1. whale behavior classifier. A minGRU model that classifies humpback dive strategies from cetacean tag data, with hierarchical annotation upstream.
  2. indysim. An individual-based simulation of larval turning under LED stimulation, with a fitted response kernel that lets a model be tested the way a real animal is measured.
  3. retrovibez. A larval reverse-crawl detection pipeline that a mechanosensation group adopted after a lab-meeting demo, with new researchers trained and install docs written.
  4. magatfairy. A one-command bridge that carries legacy MATLAB analyzer data into HDF5 and Python, with a MATLAB Engine wrapper, output validation and a CLI.

How this maps to developer experience

  1. Teach what developers can build. I have shipped CLIs, bridges and analysis pipelines, and I teach them the way I taught a lab to use retrovibez, through a live demo, trained users and written docs. The wavves method already reads as a teachable deck for an API, an SDK, plugins and automations.
  2. Explore the fringes of AI models and coding agents. wavves is my standing experiment in multi-agent orchestration across repositories, with an external planner that keeps heavy reasoning off the in-editor token budget and a recent waveset layer that dispatches waves of parallel subagents to a fresh orchestrator, and indysim pushes on simulation as a way to probe a model.
  3. Build relationships with developers and feed that back to product. retrovibez was shaped by the researchers who used it. I took in their resistance, answered it by visualizing every detected event so they could check the calls and folded their feedback into the tool.
  4. Automate your work with coding agents and increase leverage. magatfairy collapses a brittle multi-step setup into one command, gestele turns a whole codebase into a single document and wavves delegates heavy production passes to cloud agents while the audit layer stays local.
  5. Show up in the community. The public demos, the slide-ready walkthrough and the figure galleries are built to be shared and presented, which is the raw material for a meetup talk, a workshop or a thread.

Open the demos