AI made writing code nearly free. The scarce thing is judgment. Fluo reads code and measures the taste behind it.
Two builders, one radar. See where your skills fill each other's gaps and where your taste lines up.
Six axes + tags like verification-first, from any repo — a cheap prior sharpened by a model that reads the content.
See where your taste falls short of exemplary repos and get concrete upgrades.
Read-only, in-product. People whose skills fill your gaps and whose taste you share.
It keeps itself honest. The judge must cite evidence; a verifier checks the evidence actually exists in the repo. The self-improvement loop only keeps a change if it beats a held-out anchor — it can't flatter itself into confident nonsense.
Read-only and private by design. It clones and reads; it never writes to GitHub. Talent matching is in-product suggestion only — matched on technical signals only, with no personal data stored.
Don't just read one builder. See whether a whole team will ship, who to add to fill the gaps, and what getting fit wrong is costing you.
Every pairwise match, the overall composition, and the gaps to hire for — from public code.
A fleet of agents searches an opt-in index and surfaces only the mutual, intent-matched few.
Regretted hires, wasted interview loops, slow teams. Move the sliders to your reality.
Deploy a fleet of agent-finders across a read-only index. A token-disciplined cascade narrows the haystack; we contact only the opted-in, mutually-matched few — never cold outreach.
◆ Concept — the architecture we're building; figures are illustrative.
Yes — GET-only. We clone and read public code; we never write to GitHub, and we match on technical signals only (axes + taste), never name, location, or timezone. No personal data is stored beyond a derived fingerprint and a public handle.
Fair. We read what's public, and the deep read judges the actual code, not stars or follower counts. For at-work/private code, analysis is opt-in and consented. A quiet public profile isn't a verdict on you.
We don't claim to bottle it. We measure observable proxies — verification-first instinct, where effort lands, architectural restraint — and validate them against ground truth: real collaborators cohere on shared taste, and distributed ownership predicts shipping velocity (r≈−0.56). The judge cites evidence, and a verifier checks it exists in the code.
Distributed ownership (low bus factor) is the dominant validated predictor of shipping velocity, so it leads the formula, with shared values and coverage second. Bus factor needs commit history, so the public number stays honest about what it can't yet measure.
It reads revealed behavior across a whole body of work — far harder to fake than a résumé or one interview — and the evidence-grounding verifier rejects any claim it can't find in the code.
Reading your taste and matching is free. Enterprise Team Fit is a per-seat subscription. Do the math on what getting fit wrong is costing you today.