Minimal Python environment builder for data scientists

The moment someone hits this problem is usually their second week with Anaconda: they've installed everything, their machine is grinding, and they realize they're running a navigator, a base environment, and a dozen pre-installed packages they never touch — just to open a notebook. They go looking for a way to trim it down and find nothing that's actually designed for data science workflows specifically.

The gap persists because Anaconda's business model is built around the full distribution. Enterprise customers pay for the comprehensive package catalog and support contracts, so there's zero incentive to ship a lean mode that competes with their own product. The buyer (IT or procurement) cares about compliance and reproducibility; the user (the data scientist on a mid-range laptop) cares about startup time. Nobody in that chain is screaming loudly enough to change a product that enterprise accounts are already paying for.

What's broken for this specific case: users complain that conda is slow relative to pip, that mixing the two breaks things, and that the Navigator itself is the heaviest part — 'resource hungry to boot.' There's no way to say 'I need pandas, scikit-learn, and Jupyter, nothing else, and I want it to start in five seconds.' Miniconda gets you partway there but offers no guided experience, no curated project-type presets, and no conflict resolution help.

This is a business and not a feature because the problem recurs every time someone starts a new project, moves to a new machine, or onboards a new team member. Companies running data science teams on standardized hardware — often not the fastest machines — have this problem multiplied across dozens of developers. An IT manager who can shave 30 minutes off each new-hire setup day, and eliminate the 'my environment broke again' ticket, has a clear dollar figure to attach to a fix.

What to build

Ship a CLI and lightweight GUI that takes a project type selection (classification, NLP, time-series, etc.) and outputs a minimal, pre-validated conda environment YAML with only the packages that project type actually needs, plus a one-command installer that bypasses Navigator entirely and launches Jupyter directly.

Where to start

Target university data science programs and bootcamps first — instructors have the exact pain of setting up identical environments across 30 students on mismatched hardware, they're price-sensitive but vocal, and a single instructor win spreads to every student cohort they teach.

The hard part

Keeping the curated environment presets accurate as package ecosystems evolve — a preset that works cleanly today can silently break six months later when upstream packages release incompatible versions, and users will blame you, not the package authors.

How it makes money

Free for individuals up to three saved environment presets; team plan charged per seat per month for shared preset libraries, conflict-resolution logs, and environment version pinning across a team.