RATISS — compact, topological and modular approaches to artificial intelligence
I build and document experimental systems around RATISS, an evolving research ecosystem exploring compact neural cores, topological signals, temporal learning, modular architecture and verifiable computational traces.
The objective is not to imitate a large language model at a smaller scale. The objective is to investigate whether a lightweight system can acquire useful internal structure, preserve context, learn through explicit mechanisms and grow through carefully designed modules.
RATISS is developed as a research programme: each component is treated as an inspectable artifact, each result is tied to a concrete implementation, and each new capability is built incrementally.
| Research thread | Current focus |
|---|---|
| RATISS-Snn | Temporal spiking-neural experiments with a three-factor LCT learning rule, topological eligibility signals and CPU-only execution. |
| RATIS-Net | A compact neural core exploring explicit learning dynamics, contextual modulation and early language-generation pathways. |
| ODV-AEON | A modular extension layer for topological representations, structural memory, correlation traces and verifiable computation. |
| Embodied systems | The long-term application of compact contextual intelligence in robotics and resource-constrained environments. |
I work from implementation outward. Ideas are translated into code, run on contained experiments, inspected through their traces, then extended only when the previous layer is understood.
This approach keeps the research legible. It makes room for ambitious architecture while preserving an important distinction between what is implemented, what is measured, and what remains to be explored.
| Project | Role in the ecosystem | Link |
|---|---|---|
| Jonathan Evina — Portfolio | Public index of projects, research direction and RATISS-Snn experiment material. | Open repository |
| Scientist Research | Public research-facing space for the broader scientific direction. | Open repository |
| RATISS V10 — Physical Complexity Audit | A public Python framework exploring a physical-complexity framing for a P vs NP research question. | Open repository |
| RATISS AEON Agent | A public TypeScript component within the wider RATISS / AEON exploration. | Open repository |
Python · TypeScript · Spiking Neural Networks · Topological Data Analysis · Compact Neural Systems · Temporal Learning · Modular AI Architecture · Verifiable Computation
- Build small enough to understand. A mechanism should be inspectable before it is scaled.
- Keep the learning rule visible. Architecture is stronger when the source of adaptation can be located and studied.
- Use topology as structure, not decoration. Persistent form is investigated as a signal for memory, eligibility and correlation.
- Grow by composition. A capable system can be assembled from focused modules without erasing its core.
- State results honestly. Early experiments are foundations to extend, not claims to overstate.
I welcome technically serious conversations around compact AI, topological learning, spiking neural systems, embedded intelligence, robotics, open research artifacts and responsible experimental collaboration.
For research identity and public record: ORCID 0009-0000-4092-5313
Independent research in progress · Cameroon · RATISS ecosystem