About

A fund built the way an AI lab would build one.

AYVID is built on the conviction that the next great quantitative firms will not be hedge funds with an AI team. They’ll be AI systems that happen to trade. That conviction shapes every choice below.

Why now

The 1985 moment, but for language models.

In the mid-1980s, a handful of funds — Renaissance, D. E. Shaw, early Citadel — quietly began using computers to find patterns in markets. It looked silly. Their LPs were skeptical. A decade later, that work had become the most valuable strategy class in the world.

We’re at a similar inflection point with AI. The biggest funds in the world have been slow to adapt — partly because compliance won’t let them, partly because the org chart doesn’t allow it. That is the opening.

AYVID exists because we believe the alpha isn’t in bolting LLMs onto existing strategies. It’s in using LLMs to find strategies that didn’t exist before — strategies that require reading, comprehending, and reasoning across the informational firehose of the modern economy.

Engineering philosophy

Modular monolith. One repo. No microservice cosplay.

The whole stack — ingestion, NER, graph propagation, agent debate, sizing, routing, kill switch — lives in one Python + C++ repository, structured as a modular monolith. We deploy it as one binary on one box, and add scale by replicating the whole thing.

That decision is unfashionable, and it is the right one for a team this size. It removes an entire class of latency, an entire class of contract drift, and an entire class of failure where one half of the trade path doesn’t know what the other half is doing.

The boundaries we do enforce are between the deterministic execution path (Tier 1, sub-100 ms) and the LLM reasoning path (Tier 3, ~30 s). The slow tier is allowed to override the fast one — but never to be on the critical path.