Screen startups by evidence - Your deal flow, graded by evidence
by•
An AI analyst for angels, VCs, and accelerators. It reads each inbound application and grades every claim: backed by artifacts, self-reported, or unverifiable as presented — plus red flags and founder questions. Humans make the decisions.
Replies
Hi Product Hunt — I'm Vassiliy, solo builder.
AI made starting a company cheap, and it shows in every investor's inbox: more decks, more polished narratives, more specific-sounding numbers. A faster summary of an unverifiable claim is still unverifiable. The question that eats a screener's time isn't "what does this startup do" — it's "what here can I actually check?"
Evidence Gate reads one application and returns a brief:
a proceed / hold / escalate verdict (readiness for review, never an investment view)
every material claim graded — supported in the pack, self-reported, or unverifiable as presented — with the exact artifact that would verify it
red flags visible in the pack: metric inconsistencies, arithmetic breaks, unnamed "major" partners
founder questions ready to paste into your reply
New today: the same gate works from the other side of the table. If you're a founder, run your own deck through it before you submit and see what an evidence-first screener sees — then attach the missing artifact, or reword the claim, before an analyst flags it for you.
What it isn't, on purpose: not a score of how good the startup is, not a factuality checker (it can't tell you a claim is true, only whether it's verifiable and how), not a replacement for judgment. Not investment, legal, or financial advice. Humans make the decisions.
Where it's at: a live demo with two illustrative sample applications and pre-generated agent runs. I'm looking for investors and founders to run a real batch and tell me where it's wrong.
Try it → https://vassiliylakhonin.github.io/evidence-gate.html
Investors: send a batch from your current pipeline. Founders: run your own pack before you submit. I read every reply.
Tried it on a few decks and the grading actually flagged things I would have missed in a quick skim. The founder questions at the end are genuinely useful for the call.
@mervetc6u Thanks Merve — that's exactly the use I was aiming for: catch what a fast skim misses, then hand you sharper questions for the call. Since you're one of the first to actually try it: happy to run a couple of your real inbound decks myself, for free, and send back the full graded output. Just send me the decks (or a redacted version) and I'll turn them around.
Punched in a few real inbound decks and the grading felt honest, not punitive, calling out which numbers were actually backed by data and which were hand-waved. The red flag section caught a couple of things my partners and I had missed on first read.
@kardelenb989 Appreciate it, Kardelen — "honest, not punitive" is the exact line I kept editing toward. As a thank-you for being early: I'll grade a couple of your real inbound decks for free and send back the full breakdown — backed vs. hand-waved numbers, red flags, founder questions. Send me the decks and I'll run them.
how does it handle the flood of ai-generated pitch decks that all look equally polished on the surface