Scripted embodied-agent illustration. Questions and responses demonstrate the intended interaction, not a live model or scientific analysis service.
Ask. Inspect. Understand.
Meet Herror
A Fine-tuned Human Factors Domain Specific LLM
An answer should
help you ask better questions.
What does this recording support? Which method fits the question? Herror explores guidance that makes the reasoning and the evidence available for inspection.
THE IDEAHelp with the method.
Help with the method.
Keep the researcher in control.
A proposed contextual assistant for understanding and justifying human-factors analysis choices.
Domain adaptation and fine-tuning are intended research directions, to be tested against strong general-model, retrieval and tool-use baselines. The first setting is existing screen-based eye-tracking recordings, with scientific results produced by validated computation rather than generated prose.
THE ENVIRONMENTGuidance beside
Guidance beside
the research workflow.
The proposed assistant would help researchers understand fields, review data quality, choose supported methods, inspect assumptions and compare alternatives. Guidance would remain available as the dataset, selection or method changes.
Researchers would retain control of consequential mappings, parameters and transformations. Explanations would distinguish recorded measurements, method-derived quantities and psychological interpretations, including when the evidence is insufficient.
Contextual questions
Connect guidance to the research question and the inputs actually available.
Inspectable evidence
Show relevant sources, assumptions and alternatives behind a recommendation.
Controlled computation
Keep numerical operations traceable and reproducible independently of the explanation.
RESEARCH VARIABLESWhat does
What does
the assistant add?
The proposed evaluation asks about research decisions and understanding, alongside output accuracy.
THE QUESTION
Does help improve understanding?
Compare contextual guidance with documentation alone, looking at correct input preparation, method selection and the researcher's ability to explain the choice.
- What changes
- Contextual guidance versus documentation
- What we observe
- Import decisions, method selection and understanding
Does the assistant help researchers make and explain better-supported decisions?
A recommendation
with something behind it.
Supporting sources
Relevant, versioned passages should support the claim being made.
Computational record
Methods, parameters and transformations should be preserved with the result.
Limits and alternatives
Insufficient data and plausible competing interpretations should stay visible.
Herror currently consists of a research brief and plan. No model has been trained, service deployed or Studio integration implemented. The title descriptor expresses the intended direction; the embodied website agent uses scripted illustrative responses.
A useful assistant
helps you see the limits.
Herror explores whether guidance can improve both the work a researcher produces and their understanding of the decisions behind it.
Talk to the lab Next project: Behaviour Corpus