Viktor.comAn AI coworker that actually does the work
Promoted
dropoutt
Data platform that allows teams to train better models.
35 followers
Data platform that allows teams to train better models.
35 followers
Collaborative data management platform for your training data. Designed for the next generation AI models. Import, analyze, synthesize, and review your training data with your team.
honestly the collaborative review part sounds really useful, but it would help a lot if you could leave inline comments directly on specific rows or data samples. right now it kind of feels like feedback might get lost if the team is reviewing big batches at the same time. kind of like a google docs commenting vibe but for individual data points would be amazing
@derya69sb we have been considering a comment/chat feature with agents in it, just for the purpose you explained. thanks so much!
would love to see a diff view between dataset versions so we can see exactly what changed before merging into the main branch, kind of like git blame but for annotations
@diyarkelkit of course, that is one of the main features of dropoutt. we are simplifying data versioning into a web app and making every action reversible so teams can collaborate effectively and safely.
dropoutt
Training infrastructure keeps getting more complex. What was the biggest bottleneck you wanted ML teams to stop dealing with?
@aryan787544 it's correct that it gets more complex every day, but dropoutt is more on the data side of the training process. it is comparatively easy to simply view or perform exploratory analysis on other types of data, but AI training data is harder to work with and requires deeper analysis. that is the main problem we are focusing on.
@madebyroark @cagriokan That's an interesting distinction. It sounds like the real challenge isn't managing training infrastructure—it's helping teams build confidence in the data before they ever start training.
I'm curious whether, over time, you'll be measured more by reducing model failures, or by reducing the time it takes teams to trust a dataset enough to use it. Those seem like very different definitions of success.
love how clean the import flow is, feels like the team actually thought through what data scientists need instead of just shipping a generic dashboard. the collaborative review view is a nice touch too.
The collaborative review flow looks really well thought out, especially how analysis and synthesis seem to live right next to the raw data instead of buried in separate tabs. Whoever designed that clearly understands the actual pain of wrangling training datasets with a team.
The import flow felt smooth and the synthesis view actually made it easy to spot label inconsistencies across my dataset. Curious how it handles larger multi-modal batches.