heatdrop.ai

A decade-old open-source platform for running ML competitions, now gently nudging users toward its own successor.

A Python library for logging metrics, artifacts, and dataframes without needing a live Polyaxon server.

TDA is a Swing GUI and MCP server that parses, categorizes, and diagnoses thread dumps so you don't have to read raw stack traces at 2 AM.

A community-built automation framework trying to make ML benchmarking reproducible across the chaos of GPUs, containers, and constantly shifting software stacks.

Tock is an open-source conversational AI platform for teams who want to build bots without surrendering their data to a SaaS black box.

A Kubernetes operator that turns allreduce-style distributed training into a declarative YAML file, handling the messy pod orchestration so you don't have to.

A thin wrapper around seaborn and matplotlib that makes confusion matrices actually readable, with string labels and decent colormaps.

Hyperactive abstracts away the mess of hyperparameter tuning by separating what you optimize from how you optimize it.

A Docker-wrapped REST API that lets any language log to TensorBoard, not just TensorFlow.

A 2019 adversarial attack that fools text classifiers by replacing words with semantically similar synonyms—no model retraining required.

FfDL was IBM's attempt to run TensorFlow and PyTorch as a service on Kubernetes—now frozen in read-only mode.

Reconstruct what your convolutional network sees, layer by layer, then pipe it straight into TensorBoard.

A Python library that extends argparse to log experiments and parallelize hyperparameter search across GPUs or SLURM clusters without rewriting your training scripts.

The 2.x experiment tracker has been superseded by a new client built for foundation-model scale.

A Python library that watches your pandas or Spark data for distribution drift, then emails you when things go sideways.

A hyperparameter optimizer that treats your entire project history as its starting point, not a blank slate.

Picovoice built a benchmarking framework that pits cloud APIs, open-source models, and its own engines against the same audio datasets.

A CLI tool that brute-forces or heuristically searches the configuration space for NVIDIA's Triton Inference Server, then hands you a report on the trade-offs.

Kale turns a tagged notebook into a Kubeflow Pipeline without rewriting a line of Python.

A legacy Bayesian optimization package that still works if you can stomach Python 2.7 and Protocol Buffers.