Inspiration
Small businesses are the backbone of every economy, yet most SME owners have no access to financial forecasting or business intelligence. They run their businesses blind — guessing whether they can afford new inventory, hire staff, or expand. We wanted to build something that puts the power of a financial analyst + a strategy consultant into every business owner's pocket, using AI.
What it does
Fi Swarm is a multi-agent AI platform that forecasts SME cashflow. You enter your income and expenses (or import a CSV), and four LLM agents — Owner, Supplier, Customer, and Bank — collaboratively analyze your business from different angles. They produce a 12-month cashflow projection with risk alerts, confidence scoring, and a professional 9-slide PDF pitch deck ready to share with investors or stakeholders.
How we built it
React 19 + TypeScript frontend with Tailwind CSS v4 and shadcn/ui. Express + tRPC backend with Drizzle ORM on MySQL. Four AI agents powered by NVIDIA's Llama 3.3 Nemotron Super 49B model, each receiving the same business data but prompted with a distinct financial perspective. Pitch deck PDFs are rendered via Puppeteer with headless Chromium. Deployed on Railway.
Challenges we ran into
The NVIDIA API key we initially used didn't have access to the default model (returned 404), so we had to find a working model. PDF generation on Railway required cross-platform Chromium path resolution.
Accomplishments that we're proud of
A fully deployed, functional application with real AI-powered features — not a mockup or prototype. Four LLM agents collaborating in real time. Auto-generated pitch deck PDFs with SVG charts and professional design. Clean codebase with 57 passing tests. Everything works end to end: login → add transactions → run simulation → export PDF.
What we learned
How to architect a multi-agent LLM system where each agent has a distinct persona and financial perspective. How to strip out proprietary platform code and replace it with vanilla, self-hosted alternatives. How to deploy a full-stack TypeScript app with Puppeteer/Chromium on Railway.
What's next for Fi Swarm
Connect to real bank APIs for automatic transaction import. Add more agent types (Tax Agent, Market Analyst). Build a mobile-first version for SME owners who primarily use phones. Implement multi-currency support for African markets. Run longitudinal accuracy studies comparing our forecast against actual business outcomes. And open-source the agent framework so others can build their own multi-agent financial tools.
Log in or sign up for Devpost to join the conversation.