EBK is a quantitative research firm. We study how decisions get made when the outcome is unknown, then build AI systems that hold up under the same scrutiny.
The practice
Research
Five programs: evolving agent populations, placebo-grade evaluation, calibrated forecasting, real-time market data, and self-hosted inference. This is where our evaluation methods are developed and stress-tested.
ResearchApplied
Production AI built alongside the people who use it. Connectors, governed agents, and the tests that keep them honest.
Federal, state & local
American-built AI for public missions. Our government practice, built on the research and the engineering. Governed, auditable AI for federal, state and local agencies, delivered with its evaluation evidence and its code.
UEI JQTMTJ6F7PL3 · CAGE 20FZ2
EBK Public SectorEvidence
Research
Autonomous agents in an evolving population, where only agents that pass a false-discovery test at 5% may reproduce. 1,090 evolution cycles in 34 days.
Agha, CMC Senior Thesis, 2026Calibration
Expected calibration error after calibration, down from 0.252, on a held-out window split by time. Brier score fell from 0.259 to 0.211.
Market data
Prediction-market instruments tracked in a real-time order-book lake spanning five venue feeds, with timestamps aligned across sources.
Contact
Research collaborations, AI engineering engagements, and public sector work.