---
title: "Data science project management with Claude · Projectify"
description: "Data science project management for questions, experiments and model releases. Claude writes the updates and a Claude Code fleet takes the pipeline chores."
url: https://projectify.si/for/data-science
---

# Data science project management for work that rarely runs straight.

> Each question gets a card, each experiment its dataset, metric and result, and each model release the same checks. Claude writes the weekly update, and a fleet takes the tests and pipeline chores overnight.

Start free: https://app.projectify.si/signup. A month of Premium free, no card. Agents never take a seat.

## What slows a data team down.

- "Nobody remembers which run won, or why we dropped the other one." Each experiment is a task with dataset version, metric and result fields. The decision is recorded with the alternatives you ruled out.
- "People want a date for research that has no date." Claude writes the weekly update from the board: what was tried, what was learned and what comes next, so the question gets an answer even when the date cannot.
- "The engineering around the model never gets done." Leave tests, refactors and docs ready on the board. The fleet takes them overnight in their own worktrees and opens pull requests.

## Fewer chores around the model, more time with the data.

- **Takes the pipeline chores.** Unit tests, schema checks, refactors and docstrings: fleet workers on your machine take them overnight and open pull requests that never merge on their own. For example: SLT-12: 14 tests added, one failing on null tenure.
- **Keeps experiments comparable.** Fields for dataset, metric, baseline and result turn the board into a log you can sort. Table shows every run side by side. For example: Baseline AUC 0.71, boosted 0.78 on the March holdout.
- **Reads the papers and the dictionary.** Papers, data dictionaries and old analyses are read when you add them, so anyone can ask what a column means or what was tried last year. For example: Data dictionary: tenure is months since first paid invoice.
- **Releases a model the same way.** Freeze the data, evaluate, second review, model card, shadow run, switch. The workflow stops at each check for a person. For example: Churn v3: evaluation done, metrics wait for Lena’s check.

## The board it starts with

- Statuses: Question, Exploring, Experimenting, Peer review, In production, Parked
- Fields: Dataset version, Metric, Baseline, Result, Notebook or PR, Owner
- Views: Board, Table, Outline, Timeline, Dashboard
- A workflow, Model release: Freeze the training data → Evaluation on the holdout → Second reviewer checks the metrics (a person checks) → Model card from the report → Shadow run against the current model → Owner approves the switch (a person checks)

## Before the team tries it.

### Does it fit machine learning project management?

Yes. Custom fields hold the dataset, metric and result for each run, Table compares them, and decisions keep the options you ruled out.

### Can Claude see our data?

Claude sees the board, the files you add and, for the fleet, the repo on your own machine. Your warehouse stays where it is, and Enterprise can run on your own servers.

### Does it replace our experiment tracker?

No. Keep the tool that logs your runs and link each run in a field. Projectify holds the questions, decisions and work around them.

### Does the fleet need a repo?

Yes. The fleet runs in Claude Code on your machine and works on code, opening pull requests in GitHub. Analysis tasks without code are run chat by chat.

Related: [Engineering firms](https://projectify.si/for/engineering.md), [Startups](https://projectify.si/for/startups.md), [SaaS](https://projectify.si/for/saas.md), [OKRs](https://projectify.si/for/okrs.md), [Product management](https://projectify.si/for/product-management.md). Every kind of work: [For your work](https://projectify.si/for.md).