Somewhere past the mist there is a village
that studies the mind.
Sound is synthesised in the browser and stays off unless you ask for it.
Biologist by obsession, physicist by method, engineer by trade.
By day he ships production agentic AI. By night he tries to make a machine that learns the way a brain does.
The village keeps a book on everyone who passes through. His page is longer than most.
Biologist by obsession, physicist by method, engineer by trade. I study how real neurons compute — spikes, synapses, plasticity — and rebuild those tricks in silicon. By day I ship production agentic AI platforms; by night I try to make a machine that learns the way a brain does.
I came at computing from the wet side. Neurons first, transistors second — which is why I find it strange that we call a matrix multiplication a “neuron” and stop there. The day job is hard systems engineering: agentic AI platforms, Graph-RAG over enterprise knowledge, LLM memory and evaluation, C++/Go services that have to answer in under half a second. The night job is spikes, synapses, plasticity, and how little silicon you need to get real learning. Otaku in the original sense: obsessive. The long question I am actually chasing, whether a mind can be moved, has its own chapter.
Three assignments on the board. Open a scroll to read the full report.
AI-native enterprise data platform for Prudential.
Agentic AI for AR wearables — Lenskart AI Glasses.
Real-time backends for mobile gaming.
The brain runs on about 20 watts — roughly a dim light bulb — and still out-learns every model trained in a data centre. The research question: what exactly is the trick, and how much of it can silicon copy? Four stations. Each one is live. Touch everything.
Dendrites collect input, the soma sums it into a voltage, and crossing roughly −55 mV fires an all-or-nothing spike down the axon. Below threshold: nothing. Above it: a full spike, every time. Myelin isn’t insulation for looks — the signal jumps node to node instead of crawling, which is how a metre-long axon stays fast.
Two neurons never touch. A spike arrives, calcium floods in, vesicles dump neurotransmitter across a ~20 nm gap, receptors on the other side catch it. Everything a neural net casually calls a “weight” is a cartoon of this one event.
A synapse changes its own strength using only the timing of the two spikes it can see — pre-before-post strengthens it, pre-after-post weakens it. No loss function, no backward pass, no dataset. A local rule, running while the system is alive. Train a jutsu in the right order and it gets stronger; in the wrong order, weaker.
Spiking networks only spend energy when a neuron actually fires — and most neurons are silent most of the time. That isn’t an optimisation bolted on afterwards, it’s the whole architecture. Below: two honest, live-measured numbers side by side.
Ops are counted, not joules. On neuromorphic hardware energy tracks events, which is the point.
The question behind everything else on this site.
Can a mind be moved?
Consciousness onto a digital substrate. A digital footprint that stays alive, and keeps learning, after the person is gone. Or, at the far end, a mind carried into another body altogether. This is the research I study and run experiments toward. Not as science fiction, but as a chain of concrete problems: what a memory is at the level of a synapse, what has to be preserved for a pattern to still be you, and how much of that silicon can hold.
Not a file. A pattern of synaptic strengths, shaped by timing. The plasticity station upstairs is the smallest version of this question.
Which structure, at which resolution, keeps a mind continuous. The spiking-network work is a bet that the answer is dynamics, not just weights.
Silicon that runs on the brain's terms: sparse, event-driven, learning while alive. Then, possibly, something that isn't silicon at all.
If you work on any part of this, neuromorphic hardware, computational neuroscience, brain-computer interfaces, memory and continuity, the philosophy of identity, or you are simply serious about the question, I want to hear from you and I would love to contribute.
Techniques written down so they survive their author. Pull a rod to unroll.
Every skill on the résumé, filed by chakra nature. Touch a nature on the wheel.
Articles, posts, experiments and finds, pinned as they happen. New paper goes up often. Come back.
Perform the seals in order. The circle will do the rest.