Inspiration
Modern financial operations are too fragmented. Businesses juggle separate tools for invoicing, anomaly detection, transaction streaming, and billing strategy. We wanted to build a single, unified "brain" for financial ops that leverages AI not just for chat, but to actually analyze raw financial data and recommend strategies. FlowLedger AI was born to centralize these fragmented tools.
What it does
FlowLedger AI is an intelligent financial operations platform featuring:
- **Smart Invoice Processing:Upload an invoice image and Google's Gemini 2.0 Vision extracts structured data (vendor, amounts, taxes, line items) directly into JSON.
- Billing Logic Engine: Configure tiered, flat, or usage-based models, and let the AI analyze your transaction history to recommend the most optimal model for your SaaS.
- Anomaly Detection: AI scans live data streams for statistical outliers, unusual invoice timings, rate deviations, and duplicate charges.
- Real-time SSE Data Streams: A live feed of all workspace transactions natively built on Server-Sent Events.
- Multi-tenant Data Isolation: Users create distinct workspaces to fully isolate context and data streams from each other.
How we built it
We built the backend on Python/Flask to handle real-time streaming, image processing, and AI integrations. We integrated Google Gemini 2.0 Flash (via the OpenRouter API) to power all four AI modules. The frontend is built entirely using Vanilla HTML, CSS, and JS to maintain complete control over the glassmorphism design system, avoiding heavy frameworks for speed. Charts are rendered dynamically using Chart.js.
Challenges we ran into
Handling multimodal image processing for the invoice upload feature was difficult. We had to ensure images were correctly encoded to Base64 before sending them to the Vision model, while simultaneously preventing the UI from freezing during the 15-30 second inference time.
Accomplishments that we're proud of
We are incredibly proud of the custom Vanilla UI. Instead of relying on templates or Tailwind, we built a beautiful, cohesive, enterprise-grade glassmorphism interface from scratch. We are also proud of implementing true multi-workspace data isolation in a hackathon timeframe.
What we learned
We learned how powerful Server-Sent Events (SSE) can be compared to complex WebSocket setups for simple one-way live event streaming. We also learned advanced prompting techniques to force Gemini to output strictly structured JSON for our backend pipeline to ingest safely.
What's next for FlowLedger AI
Our next step is integrating Plaid APIs to pull real bank feeds directly into the SSE data streams, and building a predictive cashflow forecasting dashboard.
JUDGE INSTRUCTIONS FOR LIVE DEMO: Because our app is running via a secure LocalTunnel, you will be prompted for a "Tunnel Password" when visiting the live link.
Please enter this exact IP address as the password to view the app: 102.213.251.56
Simply click "Click to Continue" after entering it!
Log in or sign up for Devpost to join the conversation.