Fynn

Fynn analyzes your bank transactions, identifies wire fees draining your freelance income, and gives you the tools to reinvest those savings through an AI-powered portfolio manager.

Comment

Website & Githubs

Website: https://fynn-rygv.onrender.com/ Github (Render): https://github.com/Polarynx/fynnagent-hackathon Github (Development): https://github.com/jameshose1/fynnagent-hackathon

Inspiration

While attempting to pinpoint underrepresented groups in the banking market, our group continually identified freelancers and small business owners as individuals whose needs were not being met. For example, when getting paid via wire transfer, small businesses and freelancers were waiting days and losing $15 to $25 every transaction. Because these businesses survive in the margins, these fees over time are revenue killers. We wanted to build a project that not only looks at your actual transaction history, identifies where you are bleeding money on fees, and suggests a change in behavior, but rather an agentic partner that has the ability to leverage the savings it suggests into a strategic coin-based investment strategy that earns you money without any supervision.

What It Does

Fynn is an AI-powered financial agent built for freelancers and small businesses. It connects to your Capital One accounts via the Nessie API, scans your transaction history, and detects wire transfer patterns that are costing you money. It then proposes switching those payments to USDC on Solana --- a stablecoin transfer that costs fractions of a cent and settles in under 10 seconds instead of 3 business days.

Beyond payment optimization, Fynn includes a live market intelligence layer that monitors financial news in real time. When it detects a strong buy signal for a cryptocurrency based on current headlines, it proposes a medium-risk investment using your calculated investable surplus. In Supervised Mode, Fynn sends an approval request to a second device before executing any action. In Autonomous Mode, it acts immediately without interruption.

Fynn also includes a full portfolio management suite with DCA strategy execution, stop win and stop loss controls, strategy duration and interval configuration, and a standalone news trading agent that monitors the market independently of any active strategy.

How We Built It

For the frontend of Fynn we used React and Vite with a custom dark-mode design system built entirely in CSS variables. The backend is Node.js with Express and WebSockets for real-time streaming. We integrated the following APIs and services:

Capital One Nessie --- banking data and transaction history

Anthropic Claude --- AI reasoning and news sentiment analysis

Coinbase --- live crypto prices

NewsAPI --- real-time financial headlines

Solana Web3.js --- devnet USDC transactions

Auth0 --- authentication with Universal Login and MFA

Netcore --- two-device P2P communication

The agent streams its reasoning token by token via WebSocket so users can watch Fynn think in real time.

Challenges We Ran Into

The Capital One Nessie API experienced significant downtime during our build, which forced us to architect a robust fallback data system so the demo would work regardless of API availability. We also had to solve a streaming JSON parsing problem when creating Fynn's reasoning response: Claude returns structured JSON but streams it character by character, so we had to wait for the complete response before extracting the reasoning and animating it sentence by sentence. Building the two-device approval flow required careful WebSocket event deduplication to prevent double-firing across tabs and devices. The DCA engine needed ref-based state management to avoid stale closures inside \texttt{setInterval} callbacks.

Accomplishments That We're Proud Of

We are most proud of the end-to-end intelligence loop. Fynn reads banking data, reasons about it using a live language model, cross-references live news and crypto prices, and proposes a specific action with exact dollar amounts and client names. The reasoning panel streams live so judges can watch the agent think. We are also proud of the trust mode system. The distinction between Autonomous and Supervised Mode is not cosmetic --- it genuinely changes how the agent behaves and who has final say over financial decisions. Building an agent that has the ability to trade coin over intervals without a human in the loop is a feature we are incredibly bullish on.

What We Learned

We learned how to build production-grade AI agent systems that combine multiple data sources into coherent, actionable insights. We deepened our understanding of WebSocket streaming, stale closure problems in React, and the architecture required for real-time financial applications. We also learned that building for freelancers is a genuinely underserved opportunity. The tools exist to give independent workers the same financial intelligence that enterprise treasury teams have while saving them time, and nobody has built it well yet.

What's Next

As far as Fynn features, we would like to flesh out some of the other panels on the dashboard. For example, in transactions, we would like to add a budget breakdown pie chart that categorizes both personal and business expenses so that freelancers and small businesses are able to have a visualization of their spending. More ambitiously, we would like to improve our agentic trading algorithms in-house rather than simply relying on open source Claude to scrape and perform our trades.

Built With

Share this project:

Updates