Vibed a ton of small projects, mostly for my own usage, to understand what agents could do, or just to satisfy my curiosity. Most are rough around the edges, and a bunch quickly became obsolete as agents got better and I stopped using them.

Browser emulators / games

https://nesvibes.tsilva.eu/ — Browser JavaScript NES emulator in ~2.5k LOC

https://scummweb.tsilva.eu/ — Run ScummVM games directly in the browser

https://github.com/tsilva/REFramework-chill — Anti-nausea improvements for Praydog’s Resident Evil VR mod, specifically for Resident Evil 7: movement vignette, snap turning, etc. WIP.

AI / ML browser experiments

https://llame.tsilva.eu/ — Run small LLMs in the browser with WebGPU

https://aipit.tsilva.eu/ — Debate simulator: pit simulated personas against each other

https://aigrounds.tsilva.eu/ — Playgrounds for experimenting with math, AI, and ML topics. WIP; lots of slop and most are unreviewed.

https://modelviz.tsilva.eu/ — ONNX model graph visualizer

https://modelarchviz.tsilva.eu/ — View model architecture diagrams, codebase, and paper side by side. Includes a chat bot that can see selected content and select content itself. WIP.

https://embeddingviz.tsilva.eu/ — Visualize model embeddings and/or layer activations for different content with PCA, UMAP, or t-SNE. WIP and poorly tested.

https://github.com/tsilva/dlab — Workbench for deep learning experiments. Open the repo in Codex, pick a training target, e.g. maximize validation accuracy on CIFAR-10, and ask it to start a research track. Then collaborate on training runs, evaluations, sweeps, and iteration toward the goal.

Datasets / visualization

https://minariviz.tsilva.eu/ — Minari dataset visualizer: https://minari.farama.org/

https://github.com/tsilva/gymrec — Record and replay gameplay from Gymnasium environments as Hugging Face datasets. Supports stable-retro environments such as NES, SNES, Genesis, etc.

https://github.com/tsilva/youtube2datasets — Convert YouTube videos to Hugging Face datasets.

Agent / coding workflow tools

https://github.com/tsilva/runbook — Run Jupyter notebooks on Modal through a CLI, with streamed outputs.

https://github.com/tsilva/agentpong — Run coding agents in VSCode/Cursor terminals, one per Aerospace window. Agents trigger desktop notifications when done; clicking a notification sends you to the correct desktop. I haven’t used this in a while, so I’m not sure it still works.

https://github.com/tsilva/agentbox — Run agent CLIs inside Docker sandboxes. Project-level config defines sandbox access. I stopped using this once agents got better built-in approval mechanisms.

https://github.com/tsilva/agentbridge — Use an agent CLI subscription as an OpenAI-compatible API server.

https://github.com/tsilva/claudesk — Agent coordinator for Claude, built before decent agent orchestrator UIs existed. Stale and probably broken now.

Personal workflow / GTD

https://github.com/tsilva/gmail2obsidian — Flush labeled Gmail messages to Obsidian.

https://github.com/tsilva/thunkd — Quick idea-capturing mobile app built with React Native. Sends notes to Gmail, which is my main GTD inbox.

https://github.com/tsilva/capture — Quickly capture thoughts to Gmail using an Alfred shortcut, for my GTD workflow.

File / document utilities

https://dedrive.tsilva.eu/ — Client-side Google Drive duplicate finder. Lets you choose which files to keep. Poorly tested; use with caution.

https://github.com/tsilva/pdfpress — Misc PDF tools: split, merge, compress, unlock.

user:	tsilvapt
created:	June 9, 2026
karma:	2
about:	https://github.com/tsilva/