Tariff Triage

Five AI agents analyze your Canadian business in real-time, map your US tariff exposure, find cheaper Canadian suppliers, and generate a personalized trade war survival plan in PDF format.

Comment

Inspiration

The 2026 US-Canada trade war hit small businesses hardest. 25-35% tariffs on steel, lumber, auto parts, and nobody built a tool that tells a specific business owner how much it's actually costing them. TariffTriage was built to answer that.

What it does

Describe your business using the form, then five AI agents collaborate in real-time to build your survival plan:

  • Supply Chain Analyst: maps your inputs and flags US dependencies
  • Tariff Calculator: uses real CBSA tariff data to compute exact dollar exposure and margin erosion (we created our own database for this by scraping the web!)
  • Geopolitical Analyst: models escalation scenarios at 25%, 30%, 35%, 40% rates
  • Supplier Scout: finds Canadian alternatives and plots them on an interactive map
  • Strategy Architect: synthesizes a prioritized 30/60/90-day action plan

How we built it

  • Backend: FastAPI with five agents orchestrated through Backboard.io shared memory
  • AI: Gemini API powering all agents with structured JSON output
  • Data: Real CBSA tariff schedule (12K HS codes in csv with pandas incorporation)
  • Frontend: React/TypeScript with a neural network agent visualization, animated SVG trade map, and - - WebSocket live updates

Challenges we ran into

  • At first, running five agents one after another made the analysis feel too slow for a live demo. Some responses took too long, especially when multiple Gemini calls were waiting on each other. We improved this by running independent agent tasks in parallel, reducing unnecessary context passed between agents, and caching repeated tariff lookups so the final output streamed much faster (the biggest thing we found was using async functions to our advantage).
    • Our first version of the map felt heavy and unreliable because too many animated elements were updating at once. Supply lines and supplier pins were lagging or overlapping when new results arrived. We fixed this by simplifying the SVG rendering, reducing the number of animated paths, and updating map layers step by step as each agent finished, which made the visualization much smoother and easier to follow.

Accomplishments that we're proud of

  • The interactive map that we used to map out trade routes. It maps supply lines turning red at the border and green Canadian pins appearing
  • Tariff rates come from real CBSA data and our entire frontend references it which makes it cool when viewing
  • Five agents visibly collaborating as a neural network makes the product feel like actual intelligence, not a chatbot (we really wanted to simulate a CNN)

What we learned

  • Multi-agent orchestration is a completely different design pattern than single-prompt apps. The hard part is shared context, not individual agent quality.
  • Backboard.io's shared memory made coordination dramatically simpler than rolling our own.
  • Ideas get you TO THE door - execution is what gets you THROUGH THE door

What's next for Tariff Triage

  • Real-time monitoring: alerts when tariff rates change
  • Verified suppliers: actual company names and contacts, not just categories
  • ERP integration: import real purchase orders instead of describing your supply chain manually
  • Multi-business support: consultants analyzing multiple clients through one dashboard

Built With

Share this project:

Updates