RL Environments

Environments That Mirror Real-World Finance

Finance-native RL environments that mirror real deal workflows, generating world-class human data for agentic actions and elevating model reasoning, graded by the operators who run them.

Get Access See How It Works

Qofi EnvironmentM&A Diligence

Select the artifacts that apply, then work the deal as an analyst would.

Data Room

Comps Build

DCF Model

The agent works documents, models, and judgment calls in sequence, and is scored on the full trajectory, not a single output.

M&A Workflow RLStepwise DiligenceIC Reasoning

Latest Model Scores

Model A

Last evaluated Oct 12

32%

Model B

Last evaluated Oct 12

20%

Model C

Last evaluated Oct 12

10%

Select modelSelect

Run evaluation

The Platform

Everything to Build, Train, and Evaluate on RL Environments

Encode finance expertise into environments to train and evaluate models, and create the post-training data that aligns AI to real deal work.

Environment SDK

Define evals, environments, and verifiers, the tools to encode a deal workflow as a gradeable task.

Get Access

Training

Train and post-train models on live finance environments, turning expert workflows into a training signal.

Get Access

Evals

Evaluate any model against the institutional standard, operator-graded, held-out, and with real headroom.

Get Access

Data Points12M+

Financial Models4,800+

Files60k+

Bespoke Annotated Datasets500+

Qofi builds finance-native environments and real deal challenges to better understand and shape the behavior of frontier intelligences.

How It Works

Built From Real Work, Graded by Real Operators

Workflow-Faithful Tasks

Agents work the way analysts do, documents, models, and judgment calls in sequence, evaluated on the full trajectory rather than a single answer.

Operator-Graded Rewards

Reward functions are written and enforced by practitioners. Where a rubric cannot capture a judgment call, a former operator grades it.

Eval and Training Modes

The same environment runs as a held-out benchmark or a training signal, measure capability, then close the gap with the data only Qofi can produce.

Evaluation

Measure Any Model Against the Standard

Run a model through a Qofi environment and see exactly where it holds up and where it breaks, scored on the deal-level reasoning that public benchmarks miss, with results that still have headroom.

Trajectory scoringHeld-out tasksOperator rubrics

Coverage

Environments Across the Deal Spectrum

Each environment is drawn from a live workflow in the repo, and the library grows with every engagement.

Data Sources

Deal Domains

Evaluate a Model in a Qofi Environment

Experiment with Qofi RL environments today.

Get Access