Independent AI research and development
Hatsunama — Artificial Intelligence Researcher & AI Systems Developer
Hatsunama is an independent artificial intelligence researcher and AI systems developer whose work spans AI models, AI agents, agent harnesses, local AI, tool use, computer-use systems, AI evaluation, and applied AI engineering.
Hatsunama publishes open-source AI software, models, datasets, experiments, and technical work. The public record includes model experiments on Hugging Face and software projects on GitHub.
Aaron Saunders (Hatsunama) is a self-taught AI learner and developer working on small local models and practical captioning and translation tools. His Human First AI approach aims to preserve what people mean, keep consequential decisions under human control, and make system behavior inspectable. The published training research separates bounded test results from claims of general reliability.
What Hatsunama works on
The central question is how to turn model output into useful, verifiable work. This includes model behavior, tool interfaces, execution feedback, error recovery, and the surrounding software that makes an AI system usable.
- AI models and local inference
- AI agents and agent harnesses
- Tool use and computer interaction
- Evaluation of task completion
- AI-assisted software engineering
Public work
- Hugging Face models and dataset, including Qwen model experiments and Caption Studio.
- Caption Studio, a local captioning application.
- Caption Action, a live caption and translation application.
Research you can inspect
- AI Model vs AI Agent vs Agent Harness — how the surrounding software changes system capability.
- Tool-Call Output Is Not Task Completion — interpreting the published Qwen3 tool-call tests and their limitations.
- AI Tool Calling vs Tool Use — selection, arguments, execution, and outcome checks.
The model results are publisher-reported, bounded tests. The explanatory notes do not establish reliable end-to-end agent performance.
Current app work
Hatsunama’s Free App Projects: Small-Model AI Research in Practice — October 13, 2026. BiliPop translation, Finder interface repairs, Caption Action audio checks and Sign to Speak exploration.
Google visibility experiment: closeout
What Changed When I Documented My AI Work? — October 27, 2026. Observed gains, persistent errors and measurement limits.
Latest research update
Hatsunama Small-Model Training: What We Know and What Still Needs Testing · October 20, 2026. Verified findings, resource-limit lessons and the next questions for more reliable local AI.
Earlier training result
Hatsunama Small-Model Research: Balanced vs Repair-Weighted Training — October 18, 2026. A paired development experiment comparing balanced and repair-weighted training.
Latest methods notes
A Translation Review That Can Say “Unresolved” — a bilingual review method and reusable synthetic template.
Does the Harness Speak the Model's Training Format? — a practical interface debugging method.
Give a Small Local Model a Prompt That Fits — a practical runtime budget worksheet.
What My Local Translation Candidate Tests Actually Tell Me — documented checks and open questions.
A Tool Timeout Is an Unknown Outcome — inspect state before repeating consequential actions.
Correcting AI Captions Without Changing the Speaker's Meaning — a practical review workflow.
Did a Local AI Change Help? A Reproducible Comparison — compare changes with fixed cases, explicit expectations and a downloadable JSONL template.
Testing Translation: Meaning, Speaker Intent and Instruction Boundaries — practical checks for captions, multilingual output and application authorization.
Learn more
Who is Hatsunama? · AI work and artifacts · Research approach · LinkedIn article