Built by Researchers and Operators From
The Full Finance Data Stack
From managed data campaigns to proprietary datasets and RL environments, every layer frontier labs need to train and evaluate finance models, built by the operators who run the work.
Data Engine
Expert-generated reasoning data, run as managed campaigns, from task design through delivery. Qofi sources the operators, enforces the standard, and ships data that holds up under frontier evals.
Explore the Data EngineProprietary Datasets
Deal-level data the open web will never hold, sourced from institutional workflows, structured and rights-cleared, and delivered model-ready.
Explore Proprietary Datasets
RL Environments
Finance-native environments for training and evaluating agents, built from the same workflows the repo captures, graded by the operators who run them.
Explore RL EnvironmentsEvery Data Product, in One Place
RL environments, supervised and preference data, specialized domains, and evaluation sets, the full stack frontier labs train and evaluate finance models on, built by the top 5% of finance operators.
Rubric and Verifier-Based RL
Reward signals defined by detailed rubrics and programmatic verifiers, authored by finance operators, so models are graded on the reasoning a sector head would accept.
Tool-Calling RL Environments
Agents learn to call the tools analysts actually use, models, data terminals, document stores, in finance-native environments with verifiable outcomes.
Computer-Use Environments
Sandboxed environments where agents operate real software end to end, spreadsheets, research portals, deal rooms, graded on task completion.
Workflow-Faithful Tasks
Agents work the way analysts do, documents, models, and judgment calls in sequence, not isolated prompts.
Operator-Graded Rewards
Scoring rubrics written and enforced by practitioners, not crowdworkers, the standard a sector head would hold.
Eval and Training Modes
The same environment runs as a held-out benchmark or a training signal, measure a model, then improve it.