-
Spaced Repetition Trainer (Simulation)
A hand-written simulation inspired by a post from @patrickc (not a real learning product, no LLM call): ten TypeScript-quirk flashcards with Again/Hard/Good/Easy grading, an SM-2-style scheduler showing each card's ease, interval and due time on a simulated clock, sliders that retune the scheduler and recompute the whole queue live, and static explanations for cards missed twice.
Source post on X -
Terminal Mac
A scripted HTML/CSS simulation inspired by the post (no real terminal, VM, macOS or agent): a host terminal transcript beside a pretend nested desktop with a menu bar, Dock and Finder window, where a scripted cursor opens Finder, opens Notes and types a line while the terminal prints a timestamped step log; Replay (or R) restarts it, and reduced motion jumps to the final state.
Source post on X -
Agent Call Console
A scripted, clearly labelled simulation of an AI agent phone call made on your behalf: pick a preset like booking a dentist appointment or describe your own call, watch the transcript stream with a live timer and status, answer the agent when it pauses to ask you, and get a summary of the outcome and what you told it; no real calls, audio or network.
Source post on X -
Agent-in-the-Loop Optimizer
A hand-written, rule-based simulation inspired by OPAL (not OPAL, no LLM): two identical tiny networks train on the same seeded batches, and a scripted watcher rewrites one run's update rule on loss spikes, plateaus and overfitting, with live loss charts, decision regions, a timestamped edit log with code diffs and a computed comparison.
Source post on X -
GDPx Domain Re-ranker
Re-rank models on SemiAnalysis's GDPx agentic benchmark by choosing which of its 11 domains count (for example excluding the cyber, bio and alignment domains where, per SemiAnalysis, safety classifiers sandbag some models), with rank changes and a per-domain heatmap built from the published scores.
Source post on X -
Local AI: Buy vs Rent
Compare buying hardware for local AI (depreciation after resale plus electricity) against renting the same tokens from an API, with break-even hours per day, a live verdict, an hours slider and a cumulative cost chart; defaults are example numbers.
Source post on X -
Periodic Table of Finite Simple Groups
Browse the finite simple groups as a periodic table (cyclic, alternating, Lie-type families, the Tits group and all 26 sporadics), with exact orders, prime factorizations and a parameter picker that computes any family member's order live.
Source post on X -
Agent Trust Pyramid
Sort common AI agent failures into an inverted trust pyramid (codebase design, static analysis, prose, human review), get a remedy per layer and a live trust score, and export a Markdown checklist with an AGENTS.md snippet.
Source post on X -
Marble Experiment Builder
Pick a child's age and curiosity to get a concept path over a small example graph, a plan card to approve, and a playable marble-ramp, sink-or-float, magnet or shadow experiment with predict-then-explain steps.
Source post on X -
Family Cyber Shore-Up
Weekend checklist of the ten family cybersecurity steps from the post, with good/better/best tiers for passwords and 2FA, a live score and time left.
Source post on X -
macOS fm: on-device LLM
Command builder and 4,096-token budget meter for the free on-device LLM in macOS 27 Terminal (fm), with a good-fit / bad-fit task picker.
Source post on X -
Grammar Mask
Toy constrained decoder: a JSON-schema grammar masks blocked tokens, side by side with an unmasked sampler that has to retry.
Source post on X -
Spiral Frontier
The jagged AI capability frontier for eighteen tasks, re-ordered by how much data each task has.
Source post on X -
Elements 3D
Orbitable 3D periodic table of all 118 elements with search and NIST emission spectra.
Source post on X -
lrand48 Rewind
Seed srand48 from a 1997 clock and check whether a P-256 seed matches its lrand48 output fingerprint.
Source post on X -
Roadster Track Mode
Press Track Mode to deploy a stylized Roadster's active aero, with illustrative downforce readouts.
Source post on X -
Procedural Pagoda: Asuka Temple Complex
Unofficial procedural reconstruction study of an Asuka-period temple complex after Horyu-ji's western precinct: Python CSG geometry rendered in Blender, with a flyover video, six stills, a blueprint-to-render matrix, the blueprint sheet and the GLB mesh.
Source post not recorded -
Handwired, by hand
Compile a small JavaScript function into an exact pure-ReLU network with hand-set weights, no training.
Source post on X -
V3R1T4S Board
Proof-or-it-didn't-happen board of open problems agents claim to have cracked, each with a one-sitting checker.
Source post on X -
Argon Synth
Gesture synthesizer you play by dragging across a procedurally drawn prairie scene.
Source post on X -
Chokepoint Watch
Maritime ops dashboard for northbound traffic through a Bosphorus-like strait, with a point-cloud terrain map.
Source post on X -
Agent Use Cases
Filterable gallery of copy/paste prompts for autonomous agents, with guardrails built in.
Source post on X -
Routine Prompt Builder + Linter
Build and lint scheduled-agent prompts against a 7-part skeleton, from schedule to failure line.
Source post on X -
AI App Serving Cost Calculator
Turn a user count into CPUs, DRAM and dollars with editable per-user ratios to sanity-check viral serving math.
Source post on X -
EmbeddingGemma Explorer
In-browser semantic search and a 2D embedding map powered by EmbeddingGemma 2 on WebGPU/WASM, no backend.
Source post on X -
Agent Team Builder
Static builder that generates config files for a Claude Code agent team: an Opus lead plus separate workers.
Source post on X -
Minimalist Prose Linter
Fully offline prose linter that flags convoluted sentences with five plain, transparent heuristics.
Source post on X -
Karpathy Output-Format Ladder
One topic rendered four ways: plain writing, inline SVG diagram, interactive micro-explainer and storyboard.
Source post on X -
Guardrails Policy Panel
Toggle guardrail categories, severities and allow/deny rules and watch a live policy summary update.
Source post on X -
Roadster Thruster Grip
Interactive force diagram: how much downforce rear thrusters need for a ~2.7 g launch.
Source post on X -
X Newsroom Board
Found / Drafted / Posted board for an AI-newsroom workflow on X, with editable drafts and a 280-char counter.
Source post on X -
Epoch Cost-Decline Explorer
How fast do transformative-tech costs fall? Compare decline rates, with AI costs dropping ~47% per quarter.
Source post on X -
Expected Value Playground
Enter cost, probability and payoff to get live expected value, with presets like the lottery.
Source post on X -
Infomarchy Agent Desk
Omarchy wallpaper overlay of AI agent session cards with status badges, a 7-day heatmap and machine stats.
Source post not recorded -
Omarchy Site Menu
Omarchy-style site command palette with mobile layout, scroll-lock and hash jumps into in-page sections.
Source post not recorded -
Rubik's Cube Graph Solve
Scramble a 2×2 pocket cube and watch a client-side BFS solve it, framed as a search over a state graph.
Source post on X -
Omarchy Command Menu
Omarchy-style marketing mock with a site-wide command palette: hit / to fuzzy-jump across Site and Manual.
Source post not recorded -
Memory Glass Ball
Radial glass explorer for a durable agent-memory folder: search leaves, reread for a job, mark a Sunday prune.
Source post not recorded