RiskWeave

Inspiration

Financial risk rarely exists in isolation. A problem at one company, industry, or market can quickly spread through suppliers, lenders, customers, and investors, yet understanding those ripple effects often requires reading hundreds of pages of filings and connecting information across dozens of sources. We wanted to build a tool that makes systemic financial risk easy to explore visually while keeping every insight grounded in real evidence rather than AI guesses.

What it does

RiskWeave transforms financial shocks into interactive maps that show how risk spreads across companies, banks, and industries. Users can describe a scenario in natural language, such as a 20% decline in commercial real estate values or a spike in oil prices, and RiskWeave builds a structured scenario, identifies directly exposed organizations, simulates how the shock propagates through connected entities, and visualizes the results in an interactive network.

Every relationship and impact can be traced back to supporting evidence from SEC filings, financial statements, and macroeconomic data, allowing users to understand not only what is affected, but why.

How we built it

We built RiskWeave using a modern full stack architecture focused on explainability and deterministic analysis.

• Frontend: Next.js, React, TypeScript, React Flow, Tailwind CSS
• Backend: FastAPI and Python
• Database: PostgreSQL and Neo4j
• AI: Gemini API for structured document extraction, natural language scenario parsing, and evidence based explanations
• Data Sources: SEC EDGAR filings, SEC XBRL financial data, and FRED macroeconomic data

Rather than allowing the language model to generate financial conclusions, Gemini is responsible for extracting structured information and interpreting user requests while deterministic algorithms perform all calculations, propagation, and risk scoring. This keeps the analysis reproducible and transparent.

Challenges we ran into

One of our biggest challenges was balancing AI with reliability. Large language models are excellent at understanding documents, but they should not be trusted to invent financial numbers or risk scores.

To solve this, we separated responsibilities between AI and deterministic computation. Gemini extracts structured relationships from financial documents while our backend performs every calculation and propagation step using verified data. Building an explainable knowledge graph with evidence attached to every relationship also required careful schema design and data normalization across multiple public financial datasets.

Accomplishments that we're proud of

• Building an interactive financial contagion engine instead of another finance chatbot.
• Creating a graph where every meaningful relationship is backed by evidence.
• Successfully combining AI with deterministic analytics to improve transparency and trust.
• Turning complex SEC filings into an intuitive visualization that anyone can explore.
• Designing an architecture that can easily expand to additional financial scenarios in the future.

What we learned

This project reinforced that AI is most powerful when paired with traditional software engineering instead of replacing it. We learned how important explainability, reproducibility, and data provenance are in financial applications. We also gained experience integrating graph databases, public financial datasets, and LLMs into a single system that produces results users can actually trust.

What's next for RiskWeave

We plan to expand RiskWeave into a broader systemic risk platform by supporting additional financial scenarios, larger knowledge graphs, real time market and news data, portfolio level exposure analysis, and collaboration features for analysts and risk teams. We also want to improve our propagation models, introduce additional financial instruments, and continue making complex financial systems easier to understand through interactive, evidence backed visualizations.

Built With

• fastapi
• gemini
• neo4j
• nextjs
• postgresql
• python
• react
• react-flow
• tailwindcss
• typescript