Applied AI for Real World Operations

I'm working on AI Engineering solutions in my lane (logistics and transport) to build practical, responsible AI systems for transport, supply chains and operational decision making.

Start a conversation

FEATURED WORK

Featured AI Projects

Applied, evaluated systems across transport, compliance and analytics.

PythonLangGraphRAGAzure AIMulti-Agent SystemsData Analytics

Last-Mile Delivery Exception Handler

LangGraph multi-agent system using RAG, critic validation, deterministic escalation rules and prompt-injection safeguards. Evaluated across 10 end-to-end scenarios.

PythonLangGraphRAGMulti-Agent Systems

View project on GitHub →

Transport Chain of Responsibility Assistant

Multi-agent compliance prototype combining regulatory-document RAG, simulated telematics, SQL queries and live weather integration.

PythonRAGMulti-Agent SystemsData Analytics

View project on GitHub →

Conservation Analytics Framework

Modular agent architecture and interpretable modelling developed around 1.36 million IoT telemetry records.

PythonData AnalyticsMulti-Agent Systems

View project on GitHub →

Freight Cost & Logistics Optimisation

Identify cost to serve drivers, optimise carrier mix, and uncover margin leakage across lanes and service types.

DIFOT & Service Performance

Diagnose root causes of delivery failures and build practical reporting that improves reliability - not just dashboards.

Practical AI Enablement

Implement lightweight AI tools and automation that reduce manual reporting and generate actionable operational insights.

Applied Agentic AI in Transport

I design and prototype task specific AI agents that solve real operational problems in transport and supply chain environments.

Agent based systems that move beyond static dashboards.Instead of simply reporting what happened, they monitor data streams, detect emerging risks, and surface operational issues automatically - reducing manual oversight and accelerating decision making.In transport environments, this means earlier detection of compliance exposure, recurring DIFOT failures, cost drift, and service instability.

PROTOTYPE

Chain of Responsibility (CoR) Compliance Agent

Monitors scheduling and compliance signals to flag potential fatigue and mass exposure risks before breaches occur.LangChain · FAISS · RAG · Python. Real regulatory retrieval over official NHVR guidance - the driver telematics feed is simulated pending a live data source.

PROTOTYPE

Distribution LEAN Waste Multi-Agent System

A multi-agent system applying the 7 Deadly Wastes (TIMWOOD) to distribution and logistics operations - surfacing Transport (excess km's and empty running), Overprocessing (handling/touch points), Waiting, Motion, Inventory location (and network design), and Defects. Rebuilding a sophisticated waste-analysis approach originally developed for industry, using synthetic data for public demonstration. Designed as an opportunity evaluation tool for any business.

IN DEVELOPMENT

Invoice Agent

Audits invoices against contracted rates to detect cost drift and billing leakage, while tracking cost per tonne and cost per kilo by lane - surfacing inefficiencies and supporting decisions to pivot lanes when performance slips.

Applied AI in Transport - Built for Operators.

I’m developing practical agent-based systems focused on compliance, service performance, and freight cost control.I'm always open to connecting with operators exploring practical AI in transport.

Based in Australia. Working across transport and supply chain environments.

Connect on LinkedIn

Built & Shipped.

Working code, not slideware. Three projects, evolved from January's early experiments into evaluated, working systems.Multi-Agent Delivery Exception Handler
LangGraph · Python · RAG · GPT-4o · LangSmithFive specialist agents in sequence - preprocessing delivery logs, classifying exceptions, resolving against an ops playbook via RAG, drafting customer messages, and validating outputs through a critic with revision loops. Typed data views keep customer PII reachable by one agent only; a deterministic rule engine outside the LLM enforces the highest-stakes escalations.90% task completion · 100% exception ID · 100% tool call accuracy · 4.3/5 coherence across 10 end-to-end test cases.Final project, UT Austin Post-Graduate Program in AI Agents for Business Applications. View on GitHub →More projects and full source on GitHub →