Perit Marketplace turns your data and expertise into training work for frontier AI labs. Two ways in: submit data you own (audio, video, code, docs, CAD and more; 29 datasets listed so far), or build RL tasks and benchmarks on our platform using our VMs and model credits. We pitch them to labs, and approved experts scale to hundreds of tasks a week. Built by Perit (YC F26), now past $11M in revenue in 4.5 months.
I'm Kshitij, cofounder of Perit with my sister Poorva. We're both IIT Kharagpur alumni and second-time YC founders, now in YC F26.
Frontier AI labs have already trained on most of the public internet. What they need now is the data the internet doesn't have: real calls, real codebases, real work captured by the people who do it, and hard tasks written by people who know a domain inside out. A lot of that sits with startups, studios, agencies and domain experts who have no idea it's valuable.
So we opened up our marketplace. There are two ways in.
1. Own data? Submit it. Today the marketplace lists 29 datasets across 9 modalities, including 2.89M+ hours of audio, 4.74M+ hours of video and 21.1M+ lines of code. Audio, video, code, documents, images, CAD, clinical, enterprise workflows, anything.
Tell us the shape and the volume. A rough description is enough to start.
We come back with what it's worth and what clearing it would take.
We package it for the labs we work with.
2. Expert in a domain? Build RL tasks with Perit Expert Tasks. Experts write agentic tasks and benchmarks on our hosted workbench. Every task is stress-tested before we pitch it to frontier AI labs.
Simple upload. You add a task bundle and a weighted rubric, and it stays in a private workspace.
Seven quality gates. A task has to pass Build, Oracle, Do Nothing, New Seed, Difficulty Probe, Human Review and Publish before it goes live.
Calibrated difficulty. It only counts if frontier models pass it between 10% and 60% of the time.
Sandboxed and auditable. Every run is isolated, and every score traces back to the exact run that produced it.
Get paid. Experts are paid when a lab picks their task, and the best ones scale to hundreds of tasks.
To submit data, head to perit.ai/marketplace and click Submit your data. To build tasks, go to perit.ai/expert-tasks. Or drop a comment below with the data you have or the domain you know best, and our team will reach out.
Perit is a YC-backed company in the current F26 batch, and we're already working with multiple frontier AI labs. We've passed $11M in revenue. Poorva and I are both 2x YC founders, and Poorva previously worked at Google training their models. YC shared our launch here: our launch on LinkedIn
I'll be here all day answering questions and replying to every comment personally.
can contributors decide their own pricing when listing datasets on the marketplace?
Only counting tasks that frontier models pass 10 to 60 percent of the time is a sharp filter, since tasks that are too easy or impossible both teach a model nothing. That band also moves every time a new model ships. When a new model starts passing a task 90 percent of the time, does the task get retired, and does the expert who built it still get paid for past use?
Are you planning to expand the marketplace with more categories of specialized training data?
is the platform also suitable for people who want to contribute domain expertise rather than just raw datasets?
How do you make sure sensitive or private information is protected throughout the data sharing process?
Can smaller AI startups use Perit to find affordable, high quality training data for their projects?
How does Perit make it easier for independent data contributors to connect with AI labs looking for quality training data?
The seven quality gates are the part I keep coming back to — especially 'Do Nothing' and the 10–60% pass band. A task a model can pass by doing nothing, or that every frontier model already aces, tells you nothing, and most benchmarks I've seen don't filter those out. Curious how often tasks actually fail the Difficulty Probe in practice — is that the gate that kills the most submissions, or is it Human Review?
Congrats on the launch, Kshitij and Poorva! The hardest part of training frontier models right now is getting high-quality, expert-verified data at scale, and Perit Marketplace makes that a lot easier to find and buy. Excited to see where this goes!!
Respect for the emphasis on "non-annualized revenue". There's nothing wrong with reporting annualized rev but many founders are stretching reality with $XXXM annualized rev (or even worse, ARR) based on a few short-term, one-off contracts.
The marketplace angle is smart. I expect to see more companies realize the value of their data and seek out solutions like this.
About Perit Marketplace on Product Hunt
“Sell your data and RL tasks to frontier AI labs”
Perit Marketplace launched on Product Hunt on November 15th, 2026 and earned 109 upvotes and 13 comments, placing #9 on the daily leaderboard. Perit Marketplace turns your data and expertise into training work for frontier AI labs. Two ways in: submit data you own (audio, video, code, docs, CAD and more; 29 datasets listed so far), or build RL tasks and benchmarks on our platform using our VMs and model credits. We pitch them to labs, and approved experts scale to hundreds of tasks a week. Built by Perit (YC F26), now past $11M in revenue in 4.5 months.
Perit Marketplace was featured in Developer Tools (523.1k followers), Artificial Intelligence (483.8k followers) and Data (2.5k followers) on Product Hunt. Together, these topics include over 233.9k products, making this a competitive space to launch in.
Who hunted Perit Marketplace?
Perit Marketplace was hunted by Garry Tan. A “hunter” on Product Hunt is the community member who submits a product to the platform — uploading the images, the link, and tagging the makers behind it. Hunters typically write the first comment explaining why a product is worth attention, and their followers are notified the moment they post. Around 79% of featured launches on Product Hunt are self-hunted by their makers, but a well-known hunter still acts as a signal of quality to the rest of the community. See the full all-time top hunters leaderboard to discover who is shaping the Product Hunt ecosystem.
Want to see how Perit Marketplace stacked up against nearby launches in real time? Check out the live launch dashboard for upvote speed charts, proximity comparisons, and more analytics.
Hey Product Hunt 👋
I'm Kshitij, cofounder of Perit with my sister Poorva. We're both IIT Kharagpur alumni and second-time YC founders, now in YC F26.
Frontier AI labs have already trained on most of the public internet. What they need now is the data the internet doesn't have: real calls, real codebases, real work captured by the people who do it, and hard tasks written by people who know a domain inside out. A lot of that sits with startups, studios, agencies and domain experts who have no idea it's valuable.
So we opened up our marketplace. There are two ways in.
1. Own data? Submit it. Today the marketplace lists 29 datasets across 9 modalities, including 2.89M+ hours of audio, 4.74M+ hours of video and 21.1M+ lines of code. Audio, video, code, documents, images, CAD, clinical, enterprise workflows, anything.
Tell us the shape and the volume. A rough description is enough to start.
We come back with what it's worth and what clearing it would take.
We package it for the labs we work with.
2. Expert in a domain? Build RL tasks with Perit Expert Tasks. Experts write agentic tasks and benchmarks on our hosted workbench. Every task is stress-tested before we pitch it to frontier AI labs.
Simple upload. You add a task bundle and a weighted rubric, and it stays in a private workspace.
Seven quality gates. A task has to pass Build, Oracle, Do Nothing, New Seed, Difficulty Probe, Human Review and Publish before it goes live.
Calibrated difficulty. It only counts if frontier models pass it between 10% and 60% of the time.
Sandboxed and auditable. Every run is isolated, and every score traces back to the exact run that produced it.
Get paid. Experts are paid when a lab picks their task, and the best ones scale to hundreds of tasks.
To submit data, head to perit.ai/marketplace and click Submit your data. To build tasks, go to perit.ai/expert-tasks. Or drop a comment below with the data you have or the domain you know best, and our team will reach out.
Perit is a YC-backed company in the current F26 batch, and we're already working with multiple frontier AI labs. We've passed $11M in revenue. Poorva and I are both 2x YC founders, and Poorva previously worked at Google training their models. YC shared our launch here: our launch on LinkedIn
I'll be here all day answering questions and replying to every comment personally.