Agentic Automation for a US Logistics Operator
Multi-agent dispatch, voice AI, and document automation wired directly into the TMS — replacing manual email, OCR, and phone triage end-to-end.
The Challenge
Dispatchers drowning in inbound emails, rate requests, and shipping documents (PODs, BOLs, invoices). The TMS held the system-of-truth, but the human workflow around it was the bottleneck.
The Architecture
What we built.
Multi-Agent Workflow
LangChain + LangGraph system with Conversational, Executor, Router, and Monitor agents routing work across channels (email, WhatsApp, API).
Document Intelligence
OCR pipeline (Tesseract + OpenCV) with auto-classification of POD / BOL / invoice / rate-request, entity extraction, and posting directly into the TMS.
Voice + NL-SQL
Real-time voice agent (Whisper ASR + Amazon Polly TTS over SIP) plus a natural-language SQL layer so dispatchers can query shipment history conversationally.
Business Value
What changed for the customer.
- More than 70% of inbound documents processed end-to-end automatically — dispatcher workload drops dramatically.
- Routine document handling shifts from humans to agents; people focus on exceptions and high-value decisions.
- Full audit trail for compliance and exception review — every critical decision is replayable.
- System scales elastically on AWS — peak-season volume doesn't require peak-season headcount.
