I built a small tool to solve a recurring problem I’ve had while starting new projects:
it’s hard (timely, costly) to find and get market feedback from assumed customers on whether an idea is valuable before committing weeks or months of work. Conducting market research on competitors and adjacents also takes time.
Most founders (myself included) do one of the following:
-ask ChatGPT (outputs vary, tends to be polite/optimistic)
-ask feedback from friends & acquaintances (confirmation bias)
-skim Reddit threads (anecdotes)
-start building before starting sales & feedback gathering (time wasted)
I wanted a consistent, structured way to pressure-test an idea fast before putting in legwork building and gathering demand signal feedback. I also wanted to stress test the usefulness of synthetic persona feedback as an alternative to surveying tools.
What PMFScan does:
-You paste your idea or a URL to a site.
-The tool runs it through synthetic buyer persona feedback based on a model recently shown to be reliable.
It returns:
-An overall demand score and customer persona based ones, with top ones highlighted.
-Top objections & reasons the idea would fail
-Similar products out there in the market
Why not just use ChatGPT directly?
-The key to the approach is generating idea-specific personas and using vector stores to map qualitative responses into quantitative signal. This approach allows avoiding averaged-out synthetic persona response biases if you simply used chatgpt. As mentioned above, it's based on a published paper that verified the viability of the approach (https://arxiv.org/pdf/2510.08338).
Why I built it:
-I wanted to get a quick signal on whether an idea is viable, where there might be blind spots or risks, who ideal customers might be, and what businesses already exist in the space.
This is a small project, shipped quickly. If you try it, I’m interested in:
-Where the framework breaks & outputs feel wrong
-What you’d expect from a true “early PMF test”
-What would make this even more useful for you?
I built a small tool to solve a recurring problem I’ve had while starting new projects: it’s hard (timely, costly) to find and get market feedback from assumed customers on whether an idea is valuable before committing weeks or months of work. Conducting market research on competitors and adjacents also takes time.
Most founders (myself included) do one of the following: -ask ChatGPT (outputs vary, tends to be polite/optimistic) -ask feedback from friends & acquaintances (confirmation bias) -skim Reddit threads (anecdotes) -start building before starting sales & feedback gathering (time wasted)
I wanted a consistent, structured way to pressure-test an idea fast before putting in legwork building and gathering demand signal feedback. I also wanted to stress test the usefulness of synthetic persona feedback as an alternative to surveying tools.
What PMFScan does: -You paste your idea or a URL to a site. -The tool runs it through synthetic buyer persona feedback based on a model recently shown to be reliable.
It returns: -An overall demand score and customer persona based ones, with top ones highlighted. -Top objections & reasons the idea would fail -Similar products out there in the market
Why not just use ChatGPT directly? -The key to the approach is generating idea-specific personas and using vector stores to map qualitative responses into quantitative signal. This approach allows avoiding averaged-out synthetic persona response biases if you simply used chatgpt. As mentioned above, it's based on a published paper that verified the viability of the approach (https://arxiv.org/pdf/2510.08338).
Why I built it: -I wanted to get a quick signal on whether an idea is viable, where there might be blind spots or risks, who ideal customers might be, and what businesses already exist in the space.
This is a small project, shipped quickly. If you try it, I’m interested in: -Where the framework breaks & outputs feel wrong -What you’d expect from a true “early PMF test” -What would make this even more useful for you?
Link: https://pmfscan.com/
Also for funsies I added a leaderboard that shows all ideas that have been submitted and also seeded it with recent YC batch companies ;)
Happy to answer any implementation questions (LLM setup, prompt chaining, vector-store design, etc.)