D.R.O.N.E.: A Neural Network Architecture with JIT Compilation and Executable Memory
D.R.O.N.E. (Dynamic Responsive Optimized Neural Engine) is a small artificial intelligence written from scratchin C. It uses no outside libraries and no pre-trained model, and it runs offline on an ordinary CPU. Each AI builton the engine is called a drone.The design splits the work of answering into two parts. An exact core, written as ordinary C, reads everymessage, keeps track of the conversation and decides what should happen: run an action, state a thought, orjust talk. A neural brain then puts that decision into words. Its neurons are small pieces of x86-64 machinecode that the engine writes at run time. Between the two parts, a set of engine checks keeps the reply honest:it must match what was really done and computed.This paper explains how each part works and how to use it. Every mechanism is shown with a real exampletaken from the running system, and the mathematics is given where it is used. Section 4 follows one messagethrough the whole engine, step by step, and section 13 shows how to build, run, extend and train it. The projecthas only just started, so the numbers are a snapshot from October 3, 2026, after eight days and six trainingrounds, not a finished result.
Authors
- Hüseyin Teoman Deniz
- Maximum Tension
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-10-04
- DOI
- https://doi.org/10.5281/zenodo.23137562
- Primary Topic
- Artificial Intelligence Applications
- Type
- article
- Field-Weighted Citation Impact
- 0.00