How Kepler built verifiable AI for financial services with Claude

2026-04-30 · Source: Claude Blog · Field: Finance & Economics — Banking & Financial Services, Capital Markets & Investment Management, FinTech & Digital Financial Services · Depth: Advanced, medium

Summary

Kepler Finance, founded in 2025 by Vinoo Ganesh and John McRaven, has developed a verifiable AI platform for financial services that indexes over 26 million SEC filings and 14,000+ companies across 27 global markets. The platform addresses the financial industry's need for auditable and accountable reporting by integrating Claude as a reasoning and interpretation layer with deterministic infrastructure. This architecture allows analysts to ask complex questions in plain English and receive answers verifiable to the exact source document, page, and line item. Kepler found Claude superior for multi-step tasks and ambiguity flagging, using Opus 4.7 for complex reasoning and Sonnet 4.6 for high-throughput stages. The system also includes proprietary specialized models for recall, achieving 94% accuracy in mapping financial statement labels.

Key takeaway

For AI Architects and AI Product Managers building solutions in regulated industries, your focus should be on creating a "trust and verification layer" around AI models. Ensure your system can trace every output back to its source, as Kepler Finance does, by separating AI's interpretive role from deterministic, auditable computation. This approach is critical for achieving compliance and user trust, especially in environments with zero tolerance for error.

Key insights

Combining AI reasoning with deterministic infrastructure enables verifiable, auditable financial analysis.

Principles

• Verification is paramount in regulated industries.
• AI models excel at interpretation, not computation.
• Match model capabilities to specific pipeline stages.

Method

Decompose workflows into multi-stage pipelines, using Claude for reasoning and interpretation, and deterministic environments for provably correct operations like computation and fiscal period resolution, all supported by a proprietary financial ontology.

In practice

• Use Claude Opus for complex reasoning tasks.
• Employ Claude Sonnet for high-throughput, constrained tasks.
• Implement automated evaluation pipelines for AI outputs.

Topics

• Verifiable AI
• Financial Services AI
• Claude AI Platform
• Deterministic Infrastructure
• SEC Filings Analysis

Best for: AI Architect, AI Product Manager, CTO, AI Engineer, MLOps Engineer, Director of AI/ML

Related on AIssential

• AI Without Data Extraction: Building Trust‑First Infrastructure for Enterprise Decision‑Making — Enterprise AI requires verifiable, user-owned infrastructure to ensure data provenance, reproducibility, and trust, moving beyond centralized data extraction.
• Claude Opus 4.6 Thinks Smarter, xAI Joins SpaceX, AI Outperforms Doctors, Standardized AI Audits — AI advancements span corporate mergers, model capabilities, ethical oversight, and practical applications like medical diagnosis.
• Governance Is Not a Prompt — Existing model risk management frameworks are inadequate for agentic AI, necessitating a new governance paradigm focused on external, enforceable controls.
• Trust First, Tools Second: Connecting AI to Salesforce in Financial Services — Establishing a robust "trust posture" is critical for AI integration in financial services, outweighing specific tool capabilities.
• From Demos to Defensible in Financial Services Copyright & Compliance for Enterprise AI - Naveen Kumar of TD Bank — Secure AI adoption in regulated sectors requires robust governance to mitigate copyright, compliance, and data privacy risks.
• Is AI Trustworthy? — AI's biggest opportunity lies in building trust for high-stakes enterprise automation, not just in model intelligence.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Claude Blog.