Why it exists

The ergonomic gap in most data-science work isn’t modeling — it’s the run-up: setting up the data, deciding on a baseline, scaffolding a training loop, picking metrics, comparing results. Scientist-Bin treats the whole loop as a conversation and lets an agent handle the setup while a human makes the judgment calls.

What it does

• Scopes a problem from a plain-language brief and drafts an experiment plan.
• Generates the training / evaluation code, runs it, inspects the output.
• Iterates on the plan when a run fails or a metric disappoints, rather than requiring a hand-edit of the notebook.

Notes

Personal project — intentionally minimal scaffolding so the agent behaviour stays legible. The point isn’t to replace a scientist; it’s to cut the cold-start cost of trying an idea.