Research
What we are learning.
Applied, not academic. We share what we have tested and can stand behind.
DeepResearch Bench
An agent built on Atom placed second overall on DeepResearch Bench, scoring 53.70, and first on insight at 57.2, ahead of Gemini 2.5 Pro.
Our durable technical focus is entity resolution and reasoning over knowledge graphs. Retrieval finds documents that mention a name; resolution reveals the single entity behind them, and reasoning runs over the relationships between resolved entities. The work that follows lives inside that surface.