What it does: - Paste a deal memo → get scoring on 8 criteria (founder, market, traction, etc.) - Every score cites specific evidence. "Strong retention" without numbers = lower score - Compare deals side-by-side, ask follow-up questions
Demo: https://www.loom.com/share/a360b329f9e849c38c1ea70ba510d178
Tech: - Claude Sonnet 4.5 for analysis (Anthropic for nuanced judgment) - Local anonymization—company/founder names scrubbed client-side before API calls - Multi-layer QA: accuracy checker catches hallucinations, auto-retry on errors, final polish
What I learned: AI coding tools make it too easy to tinker. I'd have 3 fixes going at once, creating more bugs than I solved. Had to force myself to slow down and work methodically. Bigger lesson: I spent months tweaking in isolation instead of getting external feedback. This post is me breaking that habit.
Try it: Free tier has 20 triages + 3 deep analyses/month. I'd love feedback on whether scoring feels calibrated.