Synthetic data rooms for finance

Verified EnvironmentsForPE & BankingAnd Post-Training Loops

The Synthetic Data Company builds fully synthetic data rooms for training computer use agents on finance workflows. Our 500+ environments span the full deal lifecycle — from origination through post-close — with thousands of grounded documents per data room: financial models, memos, presentations, and compliance packages.

500+

Verifiable environments

7

Deal stages

Days → Weeks+

Task horizon

Verifiable environments: 500+
Deal stages: 7
Document types: 50+
Task horizon: Days → Weeks+

Ontology-Based Generation

Every Document Is Grounded In Facts

We generate fully synthetic data rooms by building and maintaining an ontology of facts, relations, and constraints for each synthetic company across a multi-year timeline. Every document is cross-referenced and internally consistent.

Documents per data room: 10,000+
Document types: 50+
Timeline coverage: 10+ years

Computer Use Agents

Agents Navigate Real Applications

Every environment is a full desktop — Windows or macOS — with real applications installed. Agents interact with the same tools that analysts and associates use every day. No simplified APIs. No toy interfaces.

Regulatory

Filing Systems

SEC filings, compliance submissions, regulatory forms

Spreadsheet

Microsoft Excel

Financial models, sensitivity analysis, cap tables

Email

Microsoft Outlook

Deal correspondence, internal memos, client updates

Presentations

Microsoft PowerPoint

Management decks, pitchbooks, IC presentations

Due Diligence

Virtual Data Room

Document review, Q&A workflows, access management

Market Data

Bloomberg Terminal

Comparable analysis, market data, precedent transactions

CRM

Salesforce / DealCloud

Pipeline tracking, deal management, relationship mapping

DMS

Document Management

Version control, filing, document organization

Regulatory

Filing Systems

SEC filings, compliance submissions, regulatory forms

Spreadsheet

Microsoft Excel

Financial models, sensitivity analysis, cap tables

Spreadsheet

Microsoft Excel

Financial models, sensitivity analysis, cap tables

Email

Microsoft Outlook

Deal correspondence, internal memos, client updates

Presentations

Microsoft PowerPoint

Management decks, pitchbooks, IC presentations

Due Diligence

Virtual Data Room

Document review, Q&A workflows, access management

Market Data

Bloomberg Terminal

Comparable analysis, market data, precedent transactions

CRM

Salesforce / DealCloud

Pipeline tracking, deal management, relationship mapping

DMS

Document Management

Version control, filing, document organization

Regulatory

Filing Systems

SEC filings, compliance submissions, regulatory forms

Sample Agent Trace

Agent Runtime — External

Your infrastructure

Agents run on your infrastructure or ours — connect any LLM swarm via authenticated local client.

Authenticated CUA Client · REST / gRPC

Sandbox Environment

Isolated · Snapshotted

Step 01 — Data Room

Navigate to financial folder, locate Q3 income statement

Application Layer

Microsoft Excel

Microsoft Outlook

Microsoft PowerPoint

Virtual Data Room

Microsoft Excel

Microsoft Outlook

Microsoft PowerPoint

Virtual Data Room

Bloomberg Terminal

Salesforce / DealCloud

Document Management

Filing Systems

Operating System

Windows 11 Pro · macOS Sonoma

Virtual Machine

Reproducible · Reset on demand

Deploy At Scale

Thousands of sandboxed environments. On demand.

Each environment is a fully provisioned Windows VM with 8 applications installed, a complete synthetic data room, and verifiable ground-truth. Spin up hundreds in parallel for training runs, or deploy continuously for post-training evaluation loops.

Identical environments — every VM is provisioned from the same snapshot for reproducible evaluation
Parallel execution — run hundreds of agents simultaneously across independent sandboxes
Continuous evaluation — integrate into post-training loops for ongoing model improvement

Environment Fleet

Envs Deployed

2,847

Docs Processed

1,243,600

Compute Hours

4,218

Active Now

1,849

ENV-0847running

Apps8

Docs12,400

OSWin 11

Uptime14m

ENV-1203running

Apps10

Docs8,720

OSmacOS

Uptime33m

ENV-0412ready

Apps8

Docs15,100

OSWin 11

Uptime0m

ENV-2156running

Apps10

Docs9,340

OSmacOS

Uptime8m

ENV-1891provisioning

Apps8

Docs11,800

OSWin 11

Uptime0m

ENV-0098ready

Apps10

Docs7,960

OSmacOS

Uptime0m

ENV-1547running

Apps8

Docs13,200

OSWin 11

Uptime21m

ENV-2634running

Apps10

Docs10,500

OSmacOS

Uptime35m

ENV-0331ready

Apps8

Docs6,414

OSWin 11

Uptime0m

ENV-1762running

Apps10

Docs14,512

OSWin 11

Uptime11m

ENV-2401provisioning

Apps8

Docs5,330

OSmacOS

Uptime0m

ENV-0673running

Apps10

Docs7,155

OSmacOS

Uptime18m

+ 1,000s more — identical snapshots, deployed on demand

Numbers shown are illustrative

The Scaling Law

More Environments Produce Better Models

Model performance scales with the number of diverse, verifiable long-horizon training environments, so environment count directly controls downstream capability growth.

Y-axis: Average Ranking (lower is better)

Data from Qwen 3.5 Technical Report (Alibaba, 2026). Average ranking computed across BFCL-V4, VITA-Bench, DeepPlanning, Tool-Decathlon, and MCP-Mark.

Verified EnvironmentsForPE & BankingAnd Post-Training Loops

Every Document Is Grounded In Facts

Agents Navigate Real Applications

Thousands of sandboxed environments. On demand.

More Environments Produce Better Models

The deepest environments.The longest horizons.The strongest models.

The deepest environments.
The longest horizons.
The strongest models.