Silvia Labs
Financial superintelligence for everyone.
We build the models, datasets, and evaluations required to make financial intelligence truly global and personal.
Harness engineering
The tools, retrieval, and control loop around the model. What the assistant can actually do depends on the harness at least as much as the weights.
Evaluation
Standing benchmarks and blind-graded studies built around the work a professional actually does. We open-source the datasets when the data is ours to share.
Serving and efficiency
How to run a frontier agent at production scale without a frontier bill. Routing, caching, and cost measurement grounded in production traces.
Papers
2 published
Adaptive Model Routing for Agentic Workloads
Knowing the Work Before the Agent Does It
A sub-millisecond classifier predicts how many loops an agent will run before it starts, and routes the shallow tail to a smaller, cheaper model while frontier work stays on the frontier. Trained on production traces of a financial assistant.
0.34 ms p50
Classifier inference, four to five orders of magnitude below the model call it gates.
A New Frontier in AI Tax Intelligence
How Harness Engineering Improves Factual Accuracy and Verifiability
Does retrieval tooling make AI tax guidance more accurate, and easier to check? 200 expert-level questions, 1,511 answers, 2,949 blind judgments, run twice.
2,949 blind judgments
Two rubrics per answer, graded without knowing which tool setup produced it.
Come work with us
We're a small research team building what personal finance looks like when AI actually does the work. If that's the kind of thing you want to be building, get in touch.