Kepler has unveiled a sophisticated AI architecture that achieves a remarkable 94% accuracy in extracting pertinent information from 10-K filings, a significant leap compared to the 38-46% accuracy of frontier models. This capability promises to transform how financial analysts engage with data, making sure that every output is verifiable and defensible in investment committee discussions and audits.
Architecture Designed for Finance
The unique combination of language models and deterministic code forms the backbone of Kepler's system. This architecture was recently detailed in an Anthropic customer profile, which highlighted the challenges analysts face when integrating AI into financial workflows. Traditional financial software often lacks the flexibility analysts need, while standalone language models tend to produce unverifiable data. Kepler bridges this gap by structuring the workflow—using language models to interpret questions and generate narratives, while code manages data retrieval and calculations.
Dr. John McRaven, CTO of Kepler, emphasizes the importance of this systematization: "In finance, the model can’t be the whole system. We treat it as one stage in a pipeline whose job is to hand the model exactly what it needs to succeed at exactly that stage." This systematic approach reduces the risk of errors that can derail complex financial analyses.
Enhancing Analyst Workflow
Kepler's architecture not only improves accuracy but also enhances the daily workflow for buy-side analysts. Each figure produced by the platform links directly to its source, allowing analysts to trace every number back to specific filings, pages, and line items. This transparency is key for maintaining the integrity of financial documents submitted for scrutiny. Calculations are designed to be explicit and reproducible, making sure that querying the same information multiple times yields consistent results. This end-to-end auditability simplifies compliance reviews and minimizes repetitive work.
Vinoo Ganesh, CEO of Kepler, highlights the effectiveness of the system, stating, "On our workloads, Claude was the model that consistently held the plan together. Other models would start strong and then quietly drop a constraint by step five. That behavior matters more than any benchmark score." Even a single incorrect assumption can ripple through an analysis, impacting all subsequent decisions.
Market Implications and Future Prospects
Kepler's advancements come at a critical time when the finance sector is increasingly recognizing the potential of AI. Despite other industries rapidly adopting AI technologies, finance has lagged due to stringent requirements for verification and accuracy. Kepler’s platform was built in under three months, with insights gathered from discussions with 147 financial firms. It currently serves private equity firms, hedge funds, and investment banks, with plans to expand further into private credit and adjacent regulated domains.
With a stable indexing system that encompasses 26 million SEC filings and additional public and private documents, Kepler is well-positioned to become a key player in financial AI. The backing from notable figures in AI, including the founders of OpenAI, underscores the platform's credibility and growth potential.
As the demand for verifiable AI solutions in finance rises, Kepler's architecture may set a new standard for how analysts interact with data, blending traditional financial practices with innovative AI capabilities. The future looks promising for Kepler as it continues to refine its offerings and expand its impact across the financial services sector.
