Bloom
“A Bloom filter for ‘have I told you this story?’ Probabilistic, compact, runs in the inhale before you launch into it. Sometimes silences you about anecdotes you never actually told — a feature, not a bug.”
The Problem
You are 34. You have, generously, eleven stories. You tell the one about Cher at the airport once a quarter to the same six people. They have heard it. They are too polite to say so. Your partner now mouths the punchline into her water glass at a delay of exactly one beat.
The Solution
A small daemon between your brain and your mouth. Every anecdote you tell is hashed three ways; the resulting bits get flipped in a per-listener bit array. Before any story leaves your face, we check the array. All bits set → almost certainly already told.Any bit unset → definitely fresh, proceed.
Latency
Sub-millisecond at the conversational threshold. The classifier triggers on the leading phrase — “so this is funny —”, “okay wait,”, “did I ever tell you about” — and completes lookup during the inhale. You never know it ran. You just suddenly do not have anything to say.
False-Positive Tuning
Sometimes Bloom says toldwhen you haven’t. You will, occasionally, swallow a perfectly good story about a raccoon. We are at peace with this.
Repeat Rate by Listener
Anonymized telemetry from beta cohort, normalized over a 90-day window. The cache is, predictably, worst for the people who love you most.
Pricing
TAM
The Ask
$3.5M seed. Builds the on-device hashing model, the inhale-trigger classifier, and a graceful-fail mode where the daemon murmurs “maybe” in your earpiece and you, the human, get to decide. Two engineers, one linguist, and a comedian on retainer to label training data.
Series A pitch:“We’re Datadog for the things you’ve already said.”