Bullet Tuner Studio
Status: Live.
Your muse
Can you fine tune muse glimmer to get better at writing in bullets?
Checks passed
The plan: Bullet Tuner Studio
Turn your best bullets into a training set for muse-glimmer.
Can build part of it
- Paste plain text or notes into the writer
- Pick a bullet style and generate a bullet draft
- Edit the bullets and approve the ones that sound right
Won't do: This app cannot fine tune or host muse-glimmer
See the full plan
Who it's for
People who run muse-glimmer on their own setup and want it to write clearer bullet lists.
How it works
- Paste plain text or notes into the writer.
- Pick a bullet style and generate a bullet draft.
- Edit the bullets and approve the ones that sound right.
- Add approved examples to the training set and review the pairs.
- Export the set as a JSONL file to fine tune muse-glimmer on your own setup.
Screens
- Bullet Writer. Paste text, pick a style, and generate and edit a bullet draft.
- Training Set. Review approved examples as instruction and answer pairs and remove weak ones.
- Export. Download the training set as a JSONL file and read the steps for fine tuning muse-glimmer elsewhere.
What goes in
- Text: Plain text, notes, or a short paragraph to turn into bullets, pasted in the browser.
- Text (optional): Optional examples of bullets the user likes, plus style choices such as short, detailed, or action first.
What comes out
- Text: A clean bullet list the user can edit and approve.
- File: A JSONL training file with instruction and answer pairs made from approved examples, ready for fine tuning muse-glimmer elsewhere.
Models
Model Used for Why Muse Wonder It rewrites plain text into clear bullets and turns approved examples into instruction and answer pairs for the training set. It rewrites plain text into clear bullets and turns approved examples into instruction and answer pairs for the training set. Muse Glimmer It is the target model the user wants to improve. The app prepares data for it but does not host or call it. It is the target model the user wants to improve. The app prepares data for it but does not host or call it. Limits
- This app cannot fine tune or host muse-glimmer. It builds a training file the user can use to fine tune the model on their own setup.
- Saved examples stay in the user's browser until they export them. Clearing browser data removes them.
- Bullet quality depends on the examples the user approves. Short or messy input gives weaker bullets.
- The in app preview uses muse-wonder-1.0, so it shows the target style but does not prove how a fine tuned muse-glimmer will behave.
From your inspo
The person gave no references, only a note asking to fine tune muse-glimmer so it writes better in bullets.
Takes
- The job of making bullet writing better for muse-glimmer
- A focus on bullet style, short lines, and one idea per bullet
Cost per run: about $0.02 (est.). One rewrite uses about 1,500 input tokens at $4 per million ($0.006) and about 600 output tokens at $20 per million ($0.012), for a total of $0.018, rounded to $0.02.
Recorded demo Done. Bullet Tuner Studio is built and
npm run verifyexits 0 with all steps passing: typecheck, build, start, console, smoke (5 tests), and manifest.