Independent Research & Engineering
Engineering what comes next.
We investigate the foundations of intelligent and computational systems — turning research into technologies that expand what software can achieve.
01 Research philosophy
One question, investigated from four directions.
Efficient
Computation is spent where it changes the outcome. We look for work that can be avoided, reused, or scheduled better, and we measure it honestly.
Capable
Systems that can act on underspecified goals, in complex environments, without silently guessing, and whose actions can be reviewed and reversed.
Accessible
Capability that is not limited to one hardware stack, one language, or one level of expertise.
02 Research domains
Four domains, connected by shared problems.
Core question
How can we make complex computational systems more efficient, capable, and accessible?
Every pair of domains shares at least one concrete problem. Select a domain — or a problem on the figure — to see what connects them and which projects work on it.
- Cost of computation
- Agents in tools
- Ambiguity and planning
- GPU pipelines
- Reproducible measurement
- Developer experience
03 Selected projects
Systems built to answer questions.
ZERO Research Lab
Research initiative
Investigating how much of language-model inference is necessary computation.
ROCmHub
Open-source tooling · v0.1.0
Reproducible preparation, validation, and benchmarking of open AI models for AMD ROCm.
AETHRA Engine
Game engine · Product
A real-time engine and editor built on Godot, focused on workflow, rendering research, and safe automation.
04 Current investigations
What we are trying to find out.
- Q01
Is a meaningful share of language-model inference redundant, and can it be detected cheaply enough to skip?
Evidence stage: Research hypothesisZERO Research Lab · Next: Literature taxonomy and measurement protocol
- Q02
Can an open model be taken from a model identifier to a validated run on ROCm reproducibly, with every measurement accounted for?
Evidence stage: PrototypeROCmHub · Next: First run on physical AMD hardware
- Q03
Can hardware ray-traced visibility be integrated into an existing forward renderer without regressing its baseline?
Evidence stage: PrototypeAETHRA Engine · Next: Integration into forward directional-light shading
- Q04
How should an editor expose a live scene to an automated agent so that every change is previewed, reviewable, and reversible?
Evidence stage: PrototypeAETHRA Engine · Next: Wider operation coverage under the same transaction guarantees
05 Research methodology
Measured claims, stated limits.
- 01
State the hypothesis
Write down the idea in a form that could be shown false, before building anything.
- 02
Fix the measurement
Choose metrics, baselines, hardware, and acceptable variance in advance, so results cannot be selected afterwards.
- 03
Build the instrument
Build the smallest working system that can test the idea under realistic conditions.
- 04
Report what happened
Publish the outcome with its limitations, including negative results and measurements that could not be taken.
Evidence stages
- 01 Research hypothesis A stated, falsifiable idea. No experimental evidence yet.
- 02 Active experiment Work is running to test the idea. Results are incomplete and unpublished.
- 03 Prototype Working software exists and is tested internally. Its central claims are not yet validated.
- 04 Validated result A result reproduced under a documented protocol, with the method and data available for review.
- 05 Released technology Validated work, publicly available and maintained for use by others.
Definitions and rules: How we report progress.
06 Open research
Notes now. Publications when there is something to publish.
- method How we report progress: five evidence stages Every Aporen project is labelled with one of five evidence stages, from hypothesis to released technology. This note defines them and explains why we use them.
- position One question, four domains Why Aporen's projects, which range from inference research to a game engine, are expressions of a single research direction.
-
Publications
Forthcoming
No papers have been submitted or accepted. This entry will be replaced by a list when that changes.
07Collaboration
Working on a related problem?
We are interested in hearing from researchers and engineers working on inference efficiency, agent reliability, accelerator tooling, or real-time systems — especially if you can share hardware access, data, or a hard problem.