Inspiration
The financial world often moves faster than the average investor can react. I was inspired by the collapse of major institutions like Silicon Valley Bank, where the warning signs were buried deep in hundreds of pages of SEC filings that no human has the time to read in real-time. I wanted to build a "Digital Smoke Detector" that could scan the fine print of 10-K and 10-Q filings and sound the alarm before the fire starts.
What it does
SolSignal is an automated pipeline I designed to monitor the financial health of public companies. It scrapes the latest 10-K and 10-Q filings directly from the SEC, uses the Gemini 1.5 Flash model to analyze "Risk Factors" for signs of liquidity distress, and generates an actionable Decision Signal. To ensure complete transparency, I anchored every score permanently to the Solana Blockchain, creating an immutable audit trail of corporate risk that cannot be "gaslighted" or deleted later.
How I built it
Data Ingestion: I developed a custom Python-based scraper using BeautifulSoup and Regex to bypass complex iXBRL formatting and extract narrative text for free, avoiding expensive third-party financial APIs.
AI Analysis: I utilized Gemini 1.5 Flash with a specialized "Credit Analyst" system instruction. I fed it up to 25,000 characters of risk data to generate structured JSON outputs, including Liquidity Warnings and Action Signals.
Trust Layer: I integrated the Solana web3.js library to write these signals to the Devnet ledger. I used Program Derived Addresses (PDAs) and the Memo Program to create a "Live Risk State" for each ticker.
Dynamic Vault: I built a real-time frontend that consumes the AI feed and verifies each risk assessment against the Solana transaction hash.
Challenges I ran into
Modern filings use Inline XBRL, which often breaks simple text extraction by hiding invisible HTML tags inside words like "Item 1A." I overcame this by implementing a Header-Heuristic Strategy, searching for hidden anchor IDs the SEC uses for their own internal navigation. I also managed Gemini API rate limits by implementing strategic "cooldown" periods and token-trimming to keep the project within the free-tier credits.
Accomplishments that I'm proud of
I successfully moved from "Raw Text" to "Actionable Intelligence." Seeing the system automatically flag TSLA as a "MONITOR" status while keeping MSFT as a "HOLD" based purely on deep-text analysis was a huge win for me. I am also proud of the Solana Audit Vault, which proves that AI-generated financial signals can be held accountable via blockchain immutability.
What I learned
I learned that the Context Window of an AI is its most powerful feature for finance. Being able to feed an entire "Risk Factors" section into Gemini allowed me to catch nuances. I also learned how to build a state-based registry on Solana, moving beyond simple transactions to create a "Single Source of Truth" for company risk.
Log in or sign up for Devpost to join the conversation.