HN Showcase
Back to browse
GitHub Repository

A blazingly fast, scalable metrics tracker for machine learning

2 starsPython

Aspara – Open-source ML metrics tracker that stays fast at scale

by tkng·Feb 18, 2026·2 points·1 comment
Visit ProjectView on HN

AI Analysis

●●SolidSolve My ProblemDark Horse

LTTB downsampling handles 10M metrics instantly—dashboard won't choke on scale.

Strengths
  • •LTTB-based downsampling solves real performance wall at scale
  • •Both web dashboard and Vim-keybind TUI from same data pipeline
  • •Dead simple 3-line logging API with zero local setup
Weaknesses
  • •Crowded space: MLflow, Weights & Biases, Neptune already dominate
  • •No evidence of team adoption or production deployments yet
Target Audience

ML engineers, researchers running large-scale experiments

Similar To

MLflow · Weights & Biases · Neptune

Post Description

I built Aspara, a metrics tracker for machine learning experiments that doesn't choke on large runs. Aspara applies LTTB downsampling server-side, so charts render instantly even if you have 10M data points for a single run.

Zero setup locally. Remote-ready when your team grows. Live demo (no signup): https://prednext-aspara.hf.space/

Features: - Web dashboard + Terminal UI with Vim keybindings (great over SSH) - Tag-based organization for managing hundreds of runs - Simple logging API — init(), log(), finish()

Would love feedback on what matters next: image logging or log auditing?

Similar Projects