TariffIQ

Inspiration

Global trade is the backbone of the economy, yet it relies on a dangerously complex system of HS Codes. For a small business owner, one wrong digit doesn't just cause a delay—it triggers massive customs penalties that can wipe out an entire year’s profit. We saw a gap where AI could help, but we realized standard chatbots often "hallucinate" legal data. We built TariffIQ to provide a high-precision, trustworthy co-pilot for global exporters.

What it does

TariffIQ is an AI-powered trade intelligence engine designed for speed and legal accuracy.

HS Code Classification: Users input a product description or image, and our system finds the exact 10-digit code required for customs.

Hallucination-Free Results: Unlike generic AI, we cross-reference classifications against a verified, immutable database of trade laws.

Landed Cost Calculator: We provide a real-time comparison of tariffs, taxes, and fees across major markets like India, the US, and the EU.

Explainable AI (XAI): The system provides the legal reasoning behind every code, ensuring the user is always audit-ready.

How we built it

The Development Engine: We built TariffIQ using Google Antigravity, an agentic development platform. This allowed us to deploy autonomous agents that managed the entire end-to-end development cycle—from planning our module architecture to verifying real-world trade logic across our editor and browser.

The Intelligence Layer: We leveraged Gemini API calls for the heavy lifting of semantic reasoning. When a user describes a complex product, Gemini breaks down the technical specifications to identify the likely classification.

The "Ground Truth" Data: We didn't rely on LLM training data (which can be outdated). Instead, we integrated real-time tariff and trade data directly from the WTO Database (World Trade Organization). This ensures our duty rates and trade policies are pulled from the highest global authority.

Orchestration: By combining Antigravity’s agentic workflow with direct API integrations, we created a system that doesn't just "chat" but actually executes searches, verifies data against the WTO registry, and outputs a final, compliant trade report.

Challenges we ran into

The Trust Gap: The biggest hurdle was ensuring the AI wouldn't guess a tax rate. We solved this by developing a Hybrid RAG architecture, where the LLM only interprets the product, but the "truth" (the tax rate) is pulled from our verified vector store.

Binary & OS Conflicts: Setting up the vector database locally led to significant system-level library conflicts, which we overcame by standardizing our environment and containerizing the deployment.

Data Harmonization: Merging shipping estimates, live currency rates, and tax laws into one coherent dashboard required building complex data pipelines.

Accomplishments that we're proud of

99.9% Accuracy: Successfully building a system that can accurately classify complex items (like specialized electronics) that even human experts struggle with.

Speed to Insight: Reducing a process that usually takes hours of manual legal research down to just 30 seconds.

Audit-Ready Transparency: We are incredibly proud of our "Reasoning" feature, which makes AI-driven trade legal and transparent rather than a "black box."

What we learned

We learned that in the world of Fintech and Trade, "Almost Right" is "Wrong." We shifted our focus from just making a cool UI to building a bulletproof verification layer. We also discovered the massive impact that accessible trade data can have on small businesses trying to reach global markets.