Perspective

Scientific AI needs the experiment, not just the prompt.

General AI is powerful for writing and reasoning, but experimental research requires access to raw data, metadata, algorithms, reference evidence and lab history.

Prompt

General AI usually sees only what the user pastes.

Experiment

SciPhys sees files, metadata and analysis outputs.

Memory

SciPhys can build a lab knowledge base over time.

The difference is not whether AI can write

General AI systems are excellent at language. They can explain concepts, rewrite methods sections and summarize text. For research workflows, the question is whether the AI is attached to the experiment.

If the model does not know the raw file, sample metadata, instrument settings, peak detection results, reference pattern and prior lab history, it has to infer too much from a short prompt.

Experimental research needs structured evidence

A materials researcher is not only asking for a paragraph. They are asking whether a phase assignment is credible, whether a peak is an impurity, whether a trend is systematic and what should be measured next.

Those questions require data structures: peak tables, residuals, figures, sample context, literature, reference patterns and prior experiments.

  • Raw file parsing and metadata capture.
  • Physics-aware analysis engines.
  • Reference databases and literature context.
  • Private lab memory and cross-run search.

The user experience should feel like research, not chat

Chat remains useful, but the main workspace should let researchers see the figure, adjust analysis, inspect evidence, export results and ask questions without losing context.

That is the direction of SciPhys: a scientific workspace where AI is embedded inside data analysis and lab decision-making.

What it does

Built around scientific evidence.

General AI is useful, but context-limited

A general model can help draft explanations, brainstorm hypotheses and rewrite paragraphs. The limitation is not language ability. The limitation is missing experimental state.

  • It may not know the real peak list or instrument settings.
  • It cannot inspect the uploaded archive unless the product connects it.
  • It usually does not preserve lab-specific corrections and decisions.

SciPhys starts from scientific objects

The product object in SciPhys is not a chat message. It is a run with files, metadata, analysis state, reference evidence, figures and interpretation history.

  • Domain algorithms generate measurable evidence.
  • AI explanations can cite spectrum, database, literature or lab-history evidence.
  • Outputs are designed for reports, papers and group discussion.

The moat is the feedback loop

When students and advisors review results inside the same workspace, corrections and accepted interpretations can become part of the lab's private memory.

  • More uploads create richer context.
  • More review improves future interpretation.
  • More exports turn analysis into reusable research assets.

Workflow

From raw files to research decisions.

01

Attach data

The AI reads structured experiment objects, not only pasted text.

02

Expose evidence

The answer separates measured data, references, literature and lab history.

03

Improve memory

Reviewed outcomes become reusable knowledge for future experiments.

FAQ

Questions researchers ask first.

Can researchers still use GPT or Claude with SciPhys?+

Yes. General AI remains useful for broad reasoning and writing. SciPhys focuses on the experiment-grounded layer around files, metadata, algorithms and lab memory.

What can SciPhys do that a generic chat tool usually cannot?+

SciPhys can keep raw files, metadata, analysis state, reference evidence, figures, exports and lab history connected inside one workflow.

Why does lab memory matter?+

Lab memory lets future analysis use prior samples, corrections, conclusions and project context instead of starting from an empty prompt each time.

Continue exploring

Materials AI

See the experimental AI architecture.

Platform

Explore the full workspace.

Lab data management

Build reusable lab memory.