The New Enterprise Software Category🇮🇹 Versione Italiana

PMST: Predictive Management System Transport

Beyond the legacy TMS. An embedded multi-model AI architecture designed to give B2B freight logistics a shared, predictive nervous system.

1. Why Traditional TMS Is No Longer Enough

The vast majority of Transport Management Systems (TMS) on the market today are passive relational databases born in the 1990s. They act as retrospective cash registers: telling you what was loaded yesterday, how many miles a truck ran, and what to invoice.

They cannot predict what will happen in two hours. They don't warn sales teams when promising unfeasible time windows. They don't assist dispatchers in filling linehauls prior to the 17:30 cross-dock cut-off. And they let billing departments discover proof-of-delivery (POD) damage claims only when the customer withholds payment weeks later.

2. The 3 Core Pillars of PMST

1. Upstream Process Intelligence

Shifts decision-making upstream before physical pickups: analyzes rolling 7-day pre-orders, predicts afternoon linehaul volumes by 14:00, and balances cross-dock flow ahead of evening cut-off stress.

2. Elimination of Inter-Department Friction

Sales, Dispatch, and Accounting share a unified operational intelligence. Feasibility checks, capacity constraints, and undelivered freight resolution execute in the background via AI agents, reclaiming hours of cognitive focus for human teams.

3. Local Multi-Agent Architecture & Absolute Data Privacy

No monolithic generic chatbot: a coordinated team of specialized Small/Medium Language Models (MoA) orchestrated via Model Context Protocol (MCP) on dedicated enterprise hardware. 100% industrial privacy and native B2B interoperability.

3. The Neural Network Architecture: Embedded Models & MCP Servers

PMST does not rely on a third-party black-box API. It deploys an on-premise hierarchy of domain-specific SLM/LLMs connected to live systems through open-standard Model Context Protocol (MCP) servers.

Official Technical Architecture

PMST Multi-Agent & MCP Infrastructure

⚡ Ingestion Layer: Email MCP Server (2,000+ Inbound Emails/Day)

Monitors departmental inboxes (dispatch@, sales@, customer-care@): performs instant intent triage (pickup requests, tracking queries, early closures), extracts bill-of-lading (BOL/DDT) PDFs directly to OCR, and drafts operator actions automatically.

MANAGEMENT & AUDIT MODEL

Llama 3.3 (70B-Instruct)

End-to-end holistic audit • Real customer profitability (Cost-to-Serve) • Seasonal linehaul load balancing

↓ Real-Time Inter-Agent Event Bus & MCP Protocol Synchronization ↓
Commercial Sales
Qwen 2.5 (32B)
Precise dynamic spot pricing & 7-day pre-order forecasting
Dispatch & Hub
Qwen 2.5 (14B)
Cut-off regulator, linehaul fill-rate & ZTL constraint routing
Customer Care
Mistral NeMo (12B)
Rapid exception resolution & automated pro-active ETA alerts
Billing & POD
Qwen 2.5 (32B)
Immediate POD invoice release, claim validation & EPAL balance
⚡ Specialized Process Micro-Agents & Field Sensors (Vision, Voice & Messaging)
Qwen 2.5-VL (7B)
OCR paper BOLs, stamps, and handwritten damage clauses
Whisper Turbo + Llama 3.2
Hands-free Voice-to-JSON for drivers in truck cabins
Messaging MCP Server
Interactive WhatsApp resolution with consignees
On-Premise Dedicated Server (vLLM)
2x Nvidia L40S GPUs (96GB VRAM) • Zero token costs • 100% Data Privacy
B2B Enterprise & Database MCP Gateway
Standard interfaces for Transporeon, Tesisquare, Elementum, Shippeo, project44 and SQL DBs

4. An Open Engineering Initiative

PMST is not a theoretical whitepaper drafted by academic consultants: it stems from thirty years of real frontline freight operations («Mud & Data»), grounded in the fundamental principle that technology must serve as decision support («AI suggests, it does not dictate») to restore clarity, operational velocity, and real margins to transport enterprises.

Want to learn more about the engineering philosophy behind PMST?

Read the Author's Profile →Get in Touch →