For Monday group meetings, I usually start presentation prep on Sunday afternoon. I break the group meeting presentation workflow into time blocks and write them down because I once tried "just throwing something together," and the cost was getting stumped by questions in the meeting.
18:00 | Write the conclusion first, then organize the process
Before touching the data, write one sentence: what does this week's work show? If you can't write it, it means what you did this week doesn't match the problem you were trying to solve—at that point, going back to last week's task list is more useful than forcing a PPT together.
19:00 | Data and figures
Keep only the charts and figures that support that conclusion. Get the figure standards right first: put axis units, sample size, and the meaning of error bars in the caption—don't wait to be asked. When handling chemistry or materials data, use the cheminformatics agent to run analyses (toolchains such as RDKit, Datamol, PyTDC, MedChem, and TorchDrug), and keep the scripts and versions together. Data reproducibility is the baseline—you need to be able to reproduce on the spot how a figure was made.
20:00 | Generate a first PPT draft
Use MuchSci's office assistant agent to generate an editable PPT with one sentence (the same applies to Word and Excel); editing down a first draft is much faster than building from a blank page. For occasions with a fixed structure, such as proposal and mid-term reviews, it has templates: enter the project title, research question, and preliminary ideas, and it outputs a draft covering research background, current status, methods, and technical route—you can spot missing sections at a glance.
21:00 | Give it a run-through
Present a few slides to the air. Speaking it out loud exposes two problems: which slide actually has no content, and which section is in the wrong order. When you read a script, you just slide along; when you speak, you can't fool yourself.
22:00 | Prepare three "what ifs"
Prepare one answer each for "Why did you do it this way?", "Where does the error come from?", and "What's next?" These three questions are the most likely to come up in a group meeting.
Half an hour after the meeting | Archive
Use AI transcription (listen and translate in real time, automatic segmentation and summarization, translation editing, report generation) to turn recording transcripts and key points into an archivable report; put this week's translated papers and notes into the personal knowledge base (it can import PDF, DOCX, PPTX, XLSX, TXT, and MD; personal use by default, not shared), and route handwritten key points and whiteboard photos through the personal database layer; add tasks from your advisor to next week's list.
Two boundaries to add: ask everyone present before recording—not every setting is suitable for recording; decide for yourself whether unpublished data from group meetings should be uploaded to tools—one thing I've confirmed is that uploaded papers, data, and recordings are not used for model training.
The division of labor for this graduate student AI assistant is: hand figures and first drafts to it, and keep conclusions and judgment for yourself. The time it takes to prepare for a group meeting hasn't shortened much; what has changed is the number of times you get stumped by questions in the meeting.