A private, local-first laboratory for learning, building and watching AI. It runs on my own machine: a knowledge graph of AI research, answers that arrive as moving diagrams, agents drawn as creatures, and small neural networks you can train and look inside.
A silent 100 second tour, cut from my first recorded session with the lab.
The decks
BridgeThe home deck. The brain pulses whenever the lab thinks, next to live vitals for 717 knowledge nodes.AbisaTwi for askingAsk a question and the answer is drawn step by step through the knowledge graph, with narration and a live simulation. Shown at double speed.AkontabuoTwi for reckoningThe mathematics of the field as theory cards, written large enough to walk around in. Here: attention and its variants.ChronosEighty years of AI along a golden spiral, from the McCulloch-Pitts neuron to this year's frontier releases.AtlasPapers, concepts, theories and repositories as a sky of stars, held in place by their relations.HiveAgents as creatures. Each personality slider reshapes the body, so a curious, rigorous theorist looks different from a bold scout.ForgeAssemble a network block by block and watch it learn. Every tile is one neuron, drawn as the function it computes. Shown at 1.5 times speed.ObservatoryInside a running transformer: a hand-wired induction head copies the pattern it has seen before.Simulation benchNine live simulations side by side. Here an evolutionary search fills an archive with better and more varied programs.
What I am working on now
Growing the knowledge graph from the papers I read each week, so the lab knows what I know.
A self-improvement loop: a spider-diagram benchmark scores eight capabilities, and the lab works on its weakest one, keeping a change only when it wins on held-out questions.
Voice-first Abisa: ask out loud and hear the answer while it is being drawn.
Turning the decks into learning material: one theory, one animated answer and one simulation per lesson.
The lab is private while I build it. A longer walkthrough video will follow.