N-TEK · İbrahim Alican’s independent lab · Istanbul
Every mind begins with a question.
N-TEK builds a mind that learns as it lives, and an aircraft for it to fly into the first hours after an earthquake.
The new NEXMIND is here →NOGAI Mk. II · Earthquake rapid response →NOGAI Mk. II · Earthquake rapid response
I2024Nemesis
It started with Nemesis.
A PPO agent, and our first experiments with machine learning: trial, error, reward.
It learned, and it left us a question we still chase: what would it take for a machine to learn the way we do?
II2025World models
Then it learned to dream.
We gave Nemesis a world model, RSSM and Dreamer style, so it could imagine the future before acting in it.
PonderNet let it think longer when a moment called for it; auxiliary heads made every moment teach it more.
III2025Active inference
Then we stopped paying it to learn.
Active inference became the main agent: a mind that acts to make sense of its world, not to collect a reward.
Temporal decomposition, as N-TEK defines it, let it plan across time. NEXMIND was born.
IV2026First light
The first surviving traces.
The oldest record of N-TEK that survives: temporal decomposition, already working in code.
N-TEK has built on it ever since.
V2026NOGAI Mk. I
A mind needs a body.
NOGAI Mk. I was a conventional unmanned helicopter with a shrouded tail rotor: designed as a platform for NEXMIND, not engineered to the last detail, with no exhaustive airflow studies.
It flew N-TEK’s first simulated search-and-find mission, and proved the stack in simulation.
Our vision was never the bottleneck.
VI2026CFRM
Then its failures began to teach it.
A fault in its own sensing misleads NOGAI, and it turns back to look again. The two flights are still one.
Misled once more, NOGAI learns from it with CFRM, N-TEK’s own, while it is still in the air. From here the flights part: grey without CFRM, blue with it.
The same fault meets them again: the flight with CFRM is misled once more; the one without it, twice.
Without CFRM: misled three more times in two minutes. With it: not once.
By the end, the same fault had misled the flight without CFRM 13 more times; the one with it, once.
- 1, not 13
- times the same fault misled it again, once CFRM had learned
- N-TEK simulation · one district, with and without CFRM · an earlier version, outdated
- 8/8
- sorties where CFRM learned mid-flight
- N-TEK simulation · 8 paired sorties · an earlier version, outdated
VII2026EC-AIF
Then it formed its own intermediate goals.
Event-centric active inference turns what NOGAI’s own decisions change into events. Here an event its flight control formed for itself tips the next setpoint: hold speed, not speed up.
Again its own event tips the choice: a gentle turn and more speed, where it would have held its heading and slowed.
Its search forms events of its own too. Here they steer where NOGAI searches next: toward a cell nearly 4 km from the one it would have picked.
And here they keep it on a survivor it has already detected, to locate them, instead of moving on to a new cell.
No list of subgoals. No hand-written steps. N-TEK’s contribution to active inference, out in the open.
- 18
- events of its own after four sorties, for its search and its flight
- N-TEK simulation · formed in flight · an earlier version, outdated
- 20/20
- fresh runs improved
- Paper, Table 1
NOGAI Mk. II · As designed
The air over NOGAI in cruise · N-TEK CFD
VIIITodayNOGAI Mk. II
Then we gave it wings.
NOGAI Mk. II is the compound unmanned helicopter N-TEK designs end to end: turbine-powered, in the 400 kg class.
Designed to fly faster than any unmanned helicopter in its class flies today.
An earlier build also flew in Gazebo, to check NEXMIND outside our own simulator; the aircraft model there is ours.
- 330 km/h
- max continuous speed · projected
- N-TEK design analysis · CFD and engine model
IXIn simulationThe mission
For the first hours.
NOGAI Mk. II is designed for rapid response after a major Istanbul earthquake. From a base outside the damage zone, two of them are designed to keep one searching over an area up to 100 km away, at all times.
In simulation, NEXMIND flies the whole sortie: out, the search, and back down on its pad. Hands-off.
- 755/760
- detectable survivors located · 64 fresh districts
- N-TEK simulation · an earlier version, outdated
XIn simulationThe fleet
More aircraft, less time.
Two or more NOGAIs over the same district share what each one senses and split the search between themselves.
Nobody assigns the work: the division of labour comes out of NEXMIND itself.
- 2× · 3×
- faster with two and with four
- N-TEK simulation · an earlier version, outdated
- 60/60
- missions, one to six aircraft
- N-TEK simulation · an earlier version, outdated
XIWhy
For humanity, first.
I don’t want machines that only remember. I want machines that are changed by what happens to them, the way we are.
NEXMIND learns into its own parameters as it lives: experience carried in the mind itself, not parked in a prompt.
XIINowBack the work
All we have is a creative mind, and the dedication to use it for humanity.
One person. One desktop. Imagine it with compute.
Compute first, then funding and collaboration.
Every answer is the next question.