Petrarch
Internal industrial company data for frontier labs
About Petrarch
Petrarch brings internal company data to frontier labs, starting with manufacturing and industrial companies. We source data including codebases, project files, and payments from bankruptcy courts, then de-identify and prepare the data for model training.
Public traction evidence
Each signal links to the public source used for attribution.
- X
Petrarch is building the market for specialized data locked up in enterprise companies.
Petrarch is building the market for specialized data locked up in enterprise companies. Founders @ileeilee12, @Sammycrafty, and @sudsw1234 dropped out of Harvard to unlock the next wave of training data. Backed by @ycombinator, @vivianmshen
- LinkedIn
Today, Petrarch is introducing our Constellation. We are building a community of founders, engineers, and researchers working together to democratize AI’s impact across sectors. We are now accepting… | Petrarch (YC S26) | 10 comments
We are building a community of founders, engineers, and researchers working together to democratize AI’s impact across sectors.
- LinkedIn
Petrarch YC S26 launch post
Sudhish Swain announced Petrarch (YC S26).
- X
If your company made revenue but pivoted / shutdown, let's chat.
If your company made revenue but pivoted / shutdown, let's chat. We are looking for old codebases.
- X
The data market is shifting from a farming to a mining model.
The data market is shifting from a farming to a mining model. Training the next family of long-horizon enterprise agents requires data from real, revenue-generating businesses. Link to the blog post in the comments.
- X
AI agents can write code yet fail at basic office tasks because public data rarely captures how real businesses operate.
AI agents can write code yet fail at basic office tasks because public data rarely captures how real businesses operate. Training on authentic enterprise workflows is essential for making AI dependable in real settings. Link to the blog post in the comments.
- X
Reward hacking is a fundamental challenge in building better AI models, as optimizing against imperfect signals can lead to shortcuts rather than real-world performance.
Reward hacking is a fundamental challenge in building better AI models, as optimizing against imperfect signals can lead to shortcuts rather than real-world performance. Minimizing reward hacking requires training on rich, real-world datasets that ground models in actual
- X
@yashravipati it's been a minute.
@yashravipati it's been a minute. Thanks for being our intern for the weekend (you're always welcome back 😉)