Our story
We learned to build where mistakes cost money.
Omar Garcia, our founder, has shipped production systems for thirteen years: the exchange backend behind one of Latin America’s largest crypto on-ramps, a Rust matching engine at 12,500 transactions a second, one of the first production MCP integrations.
Before Eigna he led AI engineering and built what became its base: Vesper, an open source local-first agent runtime, small experts trained on our own machines, and a learning loop with sealed tests and a release gate.
Our first client, a forecasting model for trading, shares founders with Eigna, and we say so up front. Two more follow: image and video generation for a media agency, and coding for a web development and design agency.
Our first training run cut forecast error 17.8% against its base model. It is not yet ahead of a simple last value baseline, and we publish that too.

Candidate 3
23 October, shrinkage head
HeldCandidate 2
14 October, calibrated
HeldCandidate 1
14 October, LoRA, 4 runs
Held
Case 001, our first client’s forecasting model for trading. Three candidates met the gate in October 2026. All three were held, and release 1 was not issued.







