ImportReady AI

Inspirati

Small e-commerce sellers importing goods from China (such as 1688 or Taobao) to major U.S. platforms like Amazon and eBay face significant compliance risks. Sourcing decisions are frequently made using surface-level factory quotes, while critical post-order compliance costs—ranging from mandatory lab testing to specialized dangerous goods logistics—remain hidden.
Missing a single certification can lead to port customs seizures, immediate platform delisting, or severe financial losses. However, navigating formal trade documentation is overwhelmingly complex for independent entrepreneurs. Existing generic LLM tools often hallucinate non-existent regulations or invent flawed arithmetic, while static databases fail to offer context-aware advice. ImportReady AI was created to turn opaque regulatory standards into predictable, quantifiable unit costs.

What it does

ImportReady AI is an advisory AI agent that bridges conversational product inquiry with deterministic trade compliance rules:
Natural-Language Ingestion: The user describes a product in plain language alongside preliminary sourcing parameters (e.g., target market USA, platform Amazon, quantity 300 units, purchase price $3.00/unit).
Semantic Category & Attribute Extraction: The agent identifies the item within a structured taxonomy (e.g., childrens_toys or small_consumer_electronics).
Human-in-the-Loop Clarification: When critical regulatory attributes are absent, the agent suspends execution and queries the user directly (e.g., "What is the intended age grading of this toy?" or "Does the supplier hold valid CPC or FCC documentation?").
Deterministic Risk & Cost Impact Calculation: After user confirmation, a Python-based calculation engine cross-references verified compliance datasets. For example, if a seller lacks a CPC certificate for 300 wooden blocks, the tool accounts for an estimated $600 lab testing fee, directly demonstrating an extra +$2.00/unit financial impact.
Import Readiness Report: The system compiles a clear executive summary highlighting the evaluated Risk Level (Low/Medium/High), detected missing documentation, quantified cost adjustments, and recommended next steps (such as requesting specific test reports or ordering factory samples before committing to bulk manufacturing).

How we built it

The project leverages the Strands Agents SDK running on AWS to combine language reasoning with deterministic execution:
Isolation of Reasoning and Arithmetic: Amazon Bedrock powers natural-language comprehension, candidate attribute extraction, and dynamic clarification. In contrast, all cost evaluations and regulatory matching logic are strictly isolated within deterministic Python tools to eliminate mathematical hallucinations.
Dual-Track Evidence and Rule State Engine: The underlying knowledge base adheres to strict data governance, distinguishing between evidence validity (evidence_status), formal legal status (rule_status), and product-level relevance (applicability_status). Only rules meeting both VERIFIED and EFFECTIVE criteria can trigger automated compliance burdens. Strict Workstream Isolation: Safety certifications and transport mandates are treated as independent compliance paths. For instance, consumer electronic radio-frequency clearance (FCC) is isolated from hazardous materials logistics standards (UN 38.3 / MSDS lithium battery compliance).
Separation of Accessibility and Traceability: The primary UI presents non-technical, plain-language business insights for everyday sellers, while regulatory statutes, authority sources (e.g., CPSC, eCFR, Amazon/eBay policies), and audit hashes are preserved in an expandable technical view.

Challenges we ran into

Preventing Legal and Arithmetic Hallucinations: Standard LLMs struggle with reliable unit math and frequently merge distinct regulations. Enforcing a rigid pipeline where Amazon Bedrock acts only as an orchestrator and hands all arithmetic to Python was vital.
Handling Regulatory Uncertainty & Pricing Gaps: Official testing pricing (such as Amazon-approved TIC provider verifications) is often quote-based. We had to architect the knowledge base so missing price data is flagged as QUOTE_REQUIRED rather than falsely defaulted to $0.
Balancing Technical Rigor with Usability: Presenting dense statutes from agencies like the CPSC or FCC overwhelmed non-specialist importers. We had to engineer a dual-layer reporting architecture that exposes simple executive actions upfront while preserving complete evidentiary provenance underneath.

Accomplishments that we're proud of

Functioning Human-in-the-Loop Workflow: Built an agent architecture that pauses execution to prompt users for missing critical variables instead of relying on unverified assumptions.
Actionable Unit Cost Translation: Converted abstract compliance requirements (such as ASTM F963 or battery shipping mandates) directly into transparent per-unit balance sheet impacts (e.g., +$2.00/unit).
Principled MVP Bounding: Maintained strict architectural discipline by focusing on high-risk categories for U.S. sales on Amazon and eBay, avoiding unbounded scope creep while delivering production-grade logic for targeted use cases.

What we learned

Domain Separation in Cross-Border Trade: Transport compliance (such as lithium battery UN 38.3 testing) and operational product safety (such as FCC certification) operate under completely distinct statutory authorities and must never be merged in compliance engineering.
Deterministic Tools Complement Generative Reasoning: Agentic systems are most reliable when generative models handle semantic extraction and conversational routing while deterministic code executes quantitative business logic.
Evidentiary Traceability: For automated compliance advice to be trustworthy, proposed rules, active statutes, and provider quotes must maintain clear audit trails back to official tiers of authority.

What's next for ImportReady AI

Expanded Category & Regional Coverage: Broadening beyond U.S. toys and small electronics to cover EU (CE/RoHS), UK (UKCA), and Japanese markets, as well as complex categories like cosmetics and food-contact materials.
Automated Document Verification: Introducing multimodal document analysis to allow sellers to upload factory-provided PDF test certificates and MSDS sheets for automated validation against platform requirements.
Live TIC Provider Pricing APIs: Integrating directly with verified third-party testing and certification providers to replace static planning estimates with real-time, binding cost quotes.