New code, skills and mods are posted here first. Follow on GitHub
Everything I build in the open lands here: source code, notebooks, n8n workflows, Claude skills and Claude Code mods. Every post links to the original repository or write-up. Use the categories to browse, or open the Classroom for curated collections.
I am packaging the Claude skills and Claude Code mods I use every day so other people can install them. Each one will be posted here and added to the Classroom once it is documented and safe to share.
Three synchronous n8n workflows that triage a tracked inbox, draft context-aware replies with Claude, negotiate and book meetings against live calendar availability, and write a twice-daily priority recap. The write-up covers the architecture and real execution data.
Binary classification on CelebA (202,599 face images): an ImageNet-pretrained VGG16 with a frozen base versus a three-block CNN trained from scratch. TensorFlow and Keras with Metal acceleration on Apple Silicon.
36 fine-tuning runs: BERT, DistilBERT and RoBERTa, static versus dynamic padding, IMDb and Emotion, three seeds each. DistilBERT trained about 1.94× faster than BERT for a 0.89% accuracy drop; dynamic padding saved about 4% of training time on IMDb and 11% on Emotion.
A replication of the CHI 2022 Hand Interfaces technique on Quest 2: retrieve and control eleven virtual objects by imitating their shape with your hand, carried through a migration from the Oculus Integration SDK to the Meta XR SDK.
Assigns an electric-vehicle fleet from 11 supply zones to 5 charger zones in Cagliari as a minimum-cost flow problem. Successive Shortest Path and Cycle-Canceling reach the same optimum; SSP solves the scenario in 2.7 ms against 21.6 ms.
A Docker Compose stack for real-time data: Kafka into Telegraf, InfluxDB and MongoDB for storage, Grafana dashboards, Slack alerts and Kafdrop for monitoring the topics. There is a step-by-step video walkthrough on the channel.
An integer linear program that places the fewest charging stations across Sardinian municipalities while keeping every vehicle within battery range of one, solved with Branch-and-Bound and Branch-and-Cut through IBM CPLEX.
A Dueling DQN agent that learns to play Yahtzee, with experience replay, a target network and epsilon-greedy exploration, plus scripts to train, play and plot the score distribution.
Regression analysis on the OpenML meta dataset, which describes datasets processed by learning algorithms: two preprocessing variants (missing values dropped or interpolated), standardization and linear models.
Seventeen self-contained R notebooks, from univariate and multivariate analysis through distributions, the Central Limit Theorem, confidence intervals and hypothesis tests to linear regression, each with worked examples and exercises.
Nothing here yet. The first items in this category are on the way.
What this community is, where the code lives and how to reach me.
Fine-tuning efficiency and dataset quality on small language models.
Multi-agent planning and n8n agents that do real work under human review.
Computer vision with CNNs and a reinforcement-learning agent.
A real-time streaming stack, from Kafka topics to Grafana dashboards.
Integer programming and network flows applied to EV charging in Sardinia.
Statistical inference in R and classical regression in Python.
Free-hand interaction on Quest and controller-free XR interfaces.
Reusable Claude skills for research, writing and day-to-day engineering.
The first releases are being documented. They will be announced in the community feed.
Mods that extend Claude Code with panes, status lines and guard rails.
The first releases are being documented. They will be announced in the community feed.
No collection matches your search.
This is where I share the working parts of what I build: source code, notebooks, n8n workflows, Claude skills and Claude Code mods. Every item links to the original repository or write-up, so you can read it, run it and adapt it.
It follows the same rule as the rest of my work: the machine does the preparation, a person keeps the decision, and everything in between is something you can inspect.
Everything here is free to read and learn from. Before reusing code in your own project, check the license in its repository.
Want a tutorial, found a bug, or have something to build together? Write to me or open an issue on the repository.