Projects

+ New page

Project: hutfx platform

• Created: 2026-08-16
• Status: Planned
• Owner: hutfx

Goal

hutfx is a web workspace for machine learning and science: researchers sign in and manage their ML work in one place — experiments, datasets, models, and the jobs that connect them — with a public landing page and docs site for discovery.

Scope

In (proposed — confirm before building):

• Account system with a personal workspace per user (done)
• Public landing page + /doc browser (done)
• Projects/experiments: structured record of ML runs (params, metrics, notes)
• Datasets: upload, version, link to runs
• Artifacts/files: store outputs (plots, checkpoints, logs) next to the runs
• Model registry: register versions, compare, link to runs
• Jobs: queue and monitor training/processing jobs on the server
• Public API so external programs (e.g. Python scripts) can report runs

Out (explicitly not doing, for now):

• Serving models / inference API
• In-browser notebooks (Jupyter-style)
• Team/organization accounts (single-user first)
• Billing / subscriptions

Milestones

• M1 — Experiments (MVP). The smallest useful product: a user creates a project, logs runs (name, parameters, metrics, notes) and sees them in a filterable table with a detail page. No files or compute yet.
• M2 — Datasets & artifacts. Dataset upload + artifact storage on local disk (Postgres metadata), linked to runs; image preview for plots.
• M3 — Jobs. Server-side job queue (Laravel queues) with status + log tail. Requires a decision on where jobs execute (see open questions).
• M4 — Model registry. Register model versions, compare side-by-side, download files, link back to the run that trained them.
• M5 — Public API. Token-authenticated REST endpoints for reporting runs, metrics, artifacts, and jobs from external programs.

Notes

• The feature set is proposed, not confirmed — open questions live in ../plans/2026-08-16-product-vision.md.
• Stack stays Laravel + PostgreSQL; no new infrastructure is expected before M3.
• Each milestone gets its own plan file in ../plans/ before its code is built.
• The Pipeline Engine — website spec section 3 ("Model Builder") — is described in ../plans/2026-08-17-pipelineengine.md. It is the reusable base for the website's services and processes and underpins later milestones (datasets, jobs/exports).