Interpretability for regulated finance

Black-box models, forged into glass.

Glassforge builds tools that let a bank see inside its models: what a transformer learned, why it scored an applicant the way it did, and how an agent behaves before it reaches production.

Build it, explain it, run it, measure it.

Build

Transformer Foundry

Build and tune transformer and sequence-embedding models on your own data.

Learn more
Explain

Gen AI Explainability Tools

Sparse autoencoders and Shapley attribution that turn a model's internals into evidence a validator can read.

Learn more
Run

Agent Harnesses

Controlled environments for AI agents on bank work, with a person approving every change that ships.

Learn more
Measure

Banking Benchmarks

Tests for models and agents on core banking work: credit risk, fraud, collections and model governance.

Learn more

Research

Sample posts
All posts
Shapley values, worked by hand Explainability How to write a post for this blog Guide