Tracore
Turn any document into validated, structured JSON
I've been building Workspace, an AI job marketplace. The part that ate the most time wasn't the matching, it was resume parsing. Turns out getting clean data out of random PDFs and Word docs is way harder than it sounds.
I did the obvious thing first and just asked an LLM for JSON. In production it fell apart. Same prompt, and the keys would come back nested differently. Nothing was actually validated. When something broke I couldn't reproduce it.
Then it sank in that resumes weren't special. Invoices, contracts, KYC, same headache. Companies either parse this stuff by hand, write a separate parser per format, or duct-tape an LLM and pray.
So I built the thing I wanted: you write the JSON shape you want up front, send a document, and it validates the output against that shape before handing it back. Called it Tracore. It runs in the EU because the docs I had in mind (resumes, contracts) aren't the kind you want leaving the country.
Shipped it this week. No users, no revenue, currently in the slightly terrifying "is this useful to anyone but me" phase.
Anyway, real question for people who've launched a dev tool or API here: what actually got you your first handful of users? And did free tier or free trial work better for you? I've read the generic advice, I want to hear what actually moved for you.
Happy to get into how it's built. I'll drop the link in a comment if people want it.
3 Comments
I liked the line about being in the "is this useful to anyone but me" phase.
A lot of products become harder to evaluate right after they become more ambitious.
That's an uncomfortable place to be, but usually a revealing one.
Thanks, that was the most honest line in the post. And you're right, it got harder to evaluate the moment I made it general. As a resume parser I knew if it worked; as "any document" it's fuzzy, because now it has to matter to people I haven't met. Probably the fix is getting narrow again on purpose and watching a few real users instead of guessing. Uncomfortable, but better than building blind.
That's the kind of situation I find interesting.
A conclusion can feel obvious right after a product becomes harder to evaluate.
Whether it deserves confidence is usually the harder question.
Happy to continue over email if useful.