Inspiration

Financial analysts spend a lot of time reading long financial documents such as SEC 10-K filings, risk disclosures, and management discussion sections to understand a company’s performance and risks. These documents can easily be more than 200 pages long, which makes extracting useful insights slow and difficult.

Professional investors often rely on expensive platforms like Bloomberg or FactSet to analyze this information. However, these tools are not accessible to everyone, especially students, junior analysts, and independent investors.

At the same time, many AI tools simply summarize text but do not provide reliable evidence. In finance, decisions must be traceable and supported by real data, not just generated text.

This motivated the development of FinSignal AI, a system that converts financial documents into evidence-backed signals and insights.

What it does

FinSignal AI analyzes financial documents such as:

SEC 10-K filings

risk disclosures

management discussion sections

financial statements

The system retrieves relevant information from these documents and converts it into structured insights.

It can answer questions like:

What was the company’s revenue last year?

What are the biggest risks mentioned in the latest filing?

How does one company compare with another?

Instead of giving simple summaries, FinSignal AI provides answers with citations from the original documents, allowing users to verify the information.

How we built it

First, financial documents such as SEC 10-K filings are collected and divided into sections like risk factors, management discussion, and financial statements.

We then built a hybrid retrieval system using:

BM25 keyword search to find financial terms and numbers

vector embeddings (FAISS) to capture semantic meaning

This helps the system retrieve the most relevant parts of long financial documents.

When a user asks a question, the system identifies the company and the intent of the question, retrieves the relevant evidence, and generates an answer supported by citations.

Challenges we ran into

One challenge was retrieving the correct information from very large financial documents that contain both structured tables and long narrative sections.

Another challenge was making sure the system provides reliable answers with proper evidence, instead of generating unsupported statements.

Accomplishments that we're proud of

Building a system that extracts financial insights from long documents

Generating answers that include citations to the original financial filings

Creating a hybrid retrieval system that improves search accuracy.

What we learned

We learned that building reliable AI systems requires more than just using a language model. It requires combining retrieval, reasoning, and verification so that answers can be trusted.

This project also helped us understand how AI can be used to make financial analysis more accessible and efficient.

What's next for FinSignal AI

Future improvements could include:

real-time financial news integration

better valuation models

portfolio-level analysis tools

a more advanced analyst interface.

The long-term goal is to build an affordable financial research assistant that helps more people access professional-level financial insights.

Built With

Share this project:

Updates