Founding · HYVE Labs · Summer 2022
Before the world had a word for it, the bet was already composition.
MindHYVE.ai began as a research lab in the summer of 2022 with one working thesis: a single model is incomplete, and intelligence is a composition of cooperating models. Eli is that thesis pointed at the place it is hardest to get away with being wrong.
Summer 2022
HYVE Labs
Bill Faruki starts a research lab around one intuition: an intelligent system would not be a single model, it would be a composition of cooperating models, each calibrated to a different cognitive role. HYVE is an acronym — Hyper Yielding Vectorized Entity. This is before ChatGPT exists.
November 2022
The world finds out about language models
The consumer moment proves the demand. It does not change the architectural opinion. A single model is still incomplete.
Early 2023
MindHYVE.ai
“Mind full of HYVEs.” The founder notices that his own working memory holds several parallel structures at once. The company name follows from the metaphor.
Summer 2023
Eve-Fusion
The architecture firms into a compositional fabric: a routing classifier, a domain-specialised reasoner, and frontier-grade models composed best-fit per request. The lab’s fifth working version is christened Eve-Fusion v5, and the F5/reasoner short-form follows.
2024
The portfolio takes shape
Agentic Operating Systems are scoped per discipline — Arthur for education, Chiron for healthcare, Justine for law, Theo for theology. Each runs the same fabric, calibrated to its domain. Finance is on the list from the start and is deliberately not built first.
2026
Eli
EliAI™ enters the market with its first edition, Medical Receivables Finance. It is the fifth operating system and the only one whose domain needs the other four to be good first — which is precisely why it waited.
Four anchors
The decisions that have held.
Why finance was last
Valuing a contested claim is not a finance problem wearing a finance hat. It is a medical question, then a legal question, then a counterparty question, and only then an arithmetic one. A finance product built in 2023 would have had to fake three of those four readings. Chiron, Justine and Issac had to be real before Eli could be honest.
Why the hardest asset first
The obvious first edition for a finance OS is something with clean data and a short duration. We took the opposite: receivables secured by liens against injury matters, where the payer is a settlement that has not happened, the timeline runs in years, and the evidence is a stack of clinical records. An architecture that can price this can price the easy things. The reverse is not true.
Why it refuses
The lab’s oldest position is that a system should be legible about its own limits. In education that is a tutor that says it does not know. In finance it is a valuation that declines to name a price when the cohort is thin. The discipline is the same one and it costs the same thing — a number you could have printed and did not.
What has not changed since week one
An intelligent system is a composition, not a model. That was the working sentence in 2022 and it is the working sentence today. Five cooperating models on every vertical, HYVE Labs still the named research division, and every architectural decision since descending from the same bet.