SeismiCredit

Inspiration

Credit analysts make high-stakes decisions using information scattered across SEC filings, credit agreements, earnings materials, market data, and complex Excel models. Much of their time is spent collecting documents, verifying numbers, and manually transferring information between sources and spreadsheets.

General-purpose AI tools can summarize documents, but they do not understand the full credit workflow: capital structure, debt priority, liquidity, covenants, downside risk, and recovery. After speaking with more than 25 credit professionals, we saw the need for AI infrastructure built specifically for the credit market.

What it does

SeismiCredit is an AI-native workspace for credit analysis.

It combines two core components:

1. An analyst workspace that organizes company documents, capital structure, financial performance, liquidity, and credit models in one place.
2. A credit-specific AI layer that understands credit terminology and workflows, extracts information with source traceability, and helps analysts turn documents into structured analysis.

We are starting with distressed credit, where analysis is especially complex and time-sensitive, before expanding across leveraged finance, private credit, and the broader credit market.

How we built it

We translated the workflow of a credit analyst into a modular product architecture.

The system uses AI-assisted workflows to collect and interpret financial documents, structure company and debt information, and surface the results inside an interactive analyst workspace. The workspace is designed to connect with Excel so analysts can move between AI-generated research and financial modeling without abandoning the tools they already use.

Our prototype combines a web-based workspace, Python-powered financial workflows, cloud infrastructure, large language models, and an Excel integration layer. We designed each component around traceability so analysts can review the source behind important facts and calculations.

Challenges we ran into

The hardest challenge was reliability. A plausible answer is not sufficient when an incorrect debt balance, maturity date, or claim priority can change an investment decision.

We also had to handle fragmented documents, inconsistent company and legal-entity names, changing information over time, and the difficulty of translating unstructured disclosures into structured financial models. Another major challenge was designing AI assistance that fits naturally into an analyst’s existing Excel-based workflow.

Accomplishments that we're proud of

We built an end-to-end prototype that turns fragmented credit information into an organized analytical workspace.

We are especially proud that the product is based on real analyst workflows rather than a generic document chatbot. Through more than 25 conversations with credit professionals, we refined the product around their actual needs: source-backed research, capital-structure visibility, faster modeling, and compatibility with Excel.

What we learned

We learned that credit analysts do not simply need faster document summaries. They need a system that understands how documents, legal entities, debt instruments, financial performance, and downside scenarios connect.

We also learned that trust must be part of the product architecture. Analysts need to know where a number came from, when it was reported, and how it flows into the final analysis.

What's next for SeismiCredit

Next, we will strengthen automated document acquisition, expand bidirectional Excel synchronization, and further develop and evaluate our credit-specific AI model.

We will begin with pilot users in distressed credit and use their feedback to improve the product before expanding into leveraged finance, private credit, and other parts of the credit market. Our long-term goal is to become the AI infrastructure layer for credit investing.

Built With

• ai-agents
• document-intelligence
• excel
• financial-data
• google-cloud
• large-language-models
• office.js
• python
• react