AI-Driven B2B Trade & Logistics Assistant: Market Research Automation, Compliance Verifier, and Adaptive Shipment Manager

AI-Driven B2B Trade & Logistics Assistant: Market Research Automation, Compliance Verifier, and Adaptive Shipment Manager
​Project Description
​I am looking for an experienced Full-Stack AI & Automation Developer (or small team) to build a proprietary internal web dashboard for an international wholesale trade and logistics business. The goal of this software is to streamline and automate B2B market research, verify corporate public registry listings, find correct trade commodity codes, and draft operational correspondence.
​Crucially, this system must feature continuous machine learning capabilities. The AI modules must learn from user corrections, historical data patterns, and manual overrides to steadily improve the accuracy of product classifications, email drafting, and data mapping over time.
​Core Features Required
​1. B2B Market Research Automation & Verification
​Public Directory Aggregator: Automated web workflow tools to identify potential international B2B suppliers and trading partners using public-facing trade directories and open business listings.
​Corporate Compliance Validation: Automatically verify public cross-border trade registry numbers (such as checking if a company’s public VAT or EORI registration is active using official, public government validation portals like UK Gov and EU VIES).
​2. UK Companies House Integration (OSINT)
​Real-Time Registration Checks: Direct API integration with the UK Companies House public registry to verify corporate entities.
​Activity Analysis: Automatically pull and analyze Standard Industrial Classification (SIC) codes to cross-reference and verify active business categories.
​3. AI Commodity Classification & Communication Assistant
​Intelligent Product Classification: Integrate an LLM (such as OpenAI GPT-4o or Claude 3.5 Sonnet via API) to read raw line items or commercial descriptions and determine the correct Harmonized System (HS) / Commodity Codes.
​Operational Email Drafting: An AI module that automatically drafts contextual, professional updates to suppliers, customs representatives, and freight networks based on real-time shipment milestones.
​4. Commercial Document Assistant & Logistics Integration
​Data Mapping Workflow: Securely maps parsed commercial invoice details and validated company registration data directly into structured import template fields (matching data frameworks required for UK commercial imports like IPAFFS / CHED-D structures).
​Consignment Management: API integration with major commercial carrier networks (e.g., FedEx, DHL) to automatically log package dimensions, weights, addresses, and schedule courier bookings directly from the dashboard.
​5. Adaptive Learning & Feedback Loops
​Human-in-the-Loop Review: Implement a UI mechanism allowing a human operator to review, edit, and approve AI-generated commodity codes, email drafts, and parsed data fields before they are finalized.
​Continuous Feedback Loop: The system must capture these manual human corrections and save them to a local training dataset (e.g., embeddings vector database or fine-tuning pipeline).
​Dynamic Accuracy: Future AI prompts and similarity searches must reference this historical database so the assistant dynamically learns from past adjustments, tailoring itself to the specific trade terminology and preferred writing style of the business.
​Technical Stack Preferences
​Backend: Python (FastAPI preferred) — chosen explicitly for its robust machine learning libraries, AI framework ecosystems, and data handling capabilities.
​Automation & APIs: Playwright/Selenium paired with direct API connections for public corporate databases and carrier platforms.
​AI/LLM & Learning Engine:
​OpenAI API or Anthropic API wrapper utilizing structured JSON outputs.
​Vector Database: (e.g., ChromaDB, pgvector, or Qdrant) to store historical human-corrected data for Retrieval-Augmented Generation (RAG), allowing the AI to "remember" and learn from past updates.
​UI: A simple, clean, private web dashboard for daily business operations, featuring explicit "Review, Correct & Approve" workflows.
​A Note for Developers Bidding on This Project:
​Important: The adaptive learning mechanism is a critical deliverable. The system should not simply send static prompts to an LLM. It must store a history of human overrides (e.g., if the AI proposes Commodity Code A, but the user corrects it to Commodity Code B, the system must learn to favor Code B for similar product descriptions in the future). Your proposal must outline how you plan to architect this continuous feedback and learning loop using a local database solution. All development and testing must use sandbox/test API environments and dummy mock invoices. Show More