Igloo Labs

Igloo Labs — environments for government & enterprise legacy languages

low resource gap
Languages that were mature before GitHub even before existed. Systems that power government institutions (the IRS), banks (95% of ATM transactions run on COBOL), and legacy enterprises.
Why?:

Cyber: Many bugs in government/enterprise mainframes have been undiscovered for decades. With emergent model cyber risks, securing these systems is a critical priority and requires understanding them.
Economic value: The average COBOL developer is 63, and migrations still take 4 years, 50 people, $80 million. With a retiring talent pool, the best models can help in the $91b/yr modernization industry.
Poor performance is not just explained by the severe pretraining data scarcity, but limits in model capability in understanding fragmented business logic and operational semantics from enterprise codebases.
SAP ABAP's migration deadline is Dec 31st, 2027! The 92% of Forbes Global 2000 using SAP must migrate their ABAP code or lose support, and capable models are best prepared to handle this recent incredible demand for ABAP migration.

Igloo Labs builds RL environments for frontier labs
Our researchers worked on data at Stanford/Stanford AI Labs, Meta AI, Datology, and open-source projects used by thousands of active users
Our experts spent decades on IBM mainframes, .NET, SAP ABAP, and more, where deep business logic and expertise actually lives, not on an open-source repo.
We use large legacy production codebases to create our tasks, some with 40+ years of code comments.
Environments are realistic and mimic real runtime stacks, from IBM Mainframes (still in use!), z/OS, SAP Linux, Windows Server 2012 to the modern era.
No "data slop"! Our team is research first, and our tasks are actually representative of legacy code work in the modern age.

Get in touch:

Email us - for all data samples & inquiries