Everything you need to trade smarter

TradeMate's core AI agent is powered by a LangGraph multi-step reasoning engine. It classifies your query, selects the right tools (Knowledge Graph, Document RAG, Live Rates), executes them concurrently, and synthesizes a clear, actionable answer — with citations.

• Natural language trade query understanding
• Multi-tool concurrent execution (KG + RAG + Freightos)
• Streaming responses with real-time widget rendering
• Conversation history with PostgreSQL persistence
• LLM-generated conversation titles
• Share conversations via unique public links

Powered by LangGraph StateGraph + gpt-5.4 with custom tool router

Stop manually searching through thousands of tariff schedule pages. TradeMate's Knowledge Graph contains the complete Pakistan PCT and US HTS schedules — including 4, 6, and 12-digit codes with full hierarchy traversal.

• Full Pakistan PCT coverage (scraped from TIPP)
• Complete US HTS schedule with all duty columns
• Prefix-based matching (4, 6, 12-digit codes)
• Chapter → SubChapter → Heading → SubHeading hierarchy
• Anti-dumping duties, SRO exemptions, and cess
• Linked trade procedures per HS code

Data stored in Memgraph with 340K+ relationships; queried via Cypher

TradeMate models trade data as a graph, not a flat table. Every HS code is a node with edges to its tariff rates, SRO exemptions, anti-dumping orders, regulatory procedures, cess charges, and related measures. The result is a living, queryable map of trade compliance.

• Graph-native storage in Memgraph (Bolt protocol)
• Node types: HSCode, Chapter, Heading, Tariff, Exemption, Procedure, Measure, Cess, AntiDumping
• Relationship traversal for complex compliance chains
• Real-time graph statistics (nodes, relationships, coverage)
• Admin portal for ingestion monitoring and live log streaming
• Cypher-based text search across all HS descriptions

Memgraph graph database; ingestion via custom Python scripts with AsyncIO

TradeMate integrates the Freightos API to fetch live FCL/LCL ocean and air freight spot quotes. Combined with its own cost engine, it calculates the complete DDP (Delivered Duty Paid) landed cost — factoring in THC, customs brokerage, drayage, MPF, HMF, and assessed HS duty.

• Live spot rates via Freightos API (FCL/LCL, sea/air)
• DDP landed cost engine: Inland Haulage + THC + Freight + Brokerage + Drayage + Fees + Duty
• Interactive route map with React Leaflet + CartoDB tiles
• Multi-leg route visualization (origin → transshipment → destination)
• Concurrent rate fetching (graceful Freightos fallback)
• Cost breakdown cards within the chat interface

Freightos FaaS API + custom route_engine.py DDP calculator

TradeMate doesn't just look up tariff rates — it explains them. The AI synthesizes raw tariff schedule data with regulatory context to give you a clear picture of what you'll pay, what exemptions might apply, and what compliance steps are required.

• General duty + additional duty + regulatory duty breakdown
• SRO exemption identification and applicability check
• Anti-dumping duty alerts by origin country
• Cess and additional levy identification
• Side-by-side PK vs US tariff comparison
• Plain-language compliance explanations from the AI agent

Combines Knowledge Graph Cypher queries + RAG document retrieval for regulatory context

TradeMate's voice interface uses the OpenAI Realtime API (gpt-5.4 Realtime) to enable natural, spoken trade consultations. Up to 60-second real-time voice conversations give you hands-free access to the full platform intelligence.

• OpenAI gpt-5.4 Realtime API integration
• Up to 60-second voice conversations
• Full access to trade knowledge during voice session
• Automatic transcription and response display
• Seamless handoff from voice to text conversation

OpenAI Realtime API with WebSockets; handled in server/routes/voice.py

TradeMate's RAG pipeline ingests trade policy documents, regulatory notices, and research reports into a Pinecone vector store. An automated researcher fetches and analyzes live news and trade publications hourly, keeping the knowledge base continuously updated.

• PDF and document ingestion with semantic chunking
• OpenAI text-embedding-3-small vector embeddings
• Pinecone vector store with metadata filtering
• Automated hourly research runner (news + trade publications)
• S3 persistence for ingestion checkpoints and results
• Admin portal with real-time job monitoring and drag-and-drop upload

Semantic chunking + batch Pinecone upserts; scheduler via Celery Beat