Publications
Weather-routing studies computed on ERA5 reanalysis with the per-case numbers published, plus earlier writing on maritime law, tanker operations and local AI. Older articles on general AI and quantitative trading are kept in the archive.
A full year of weather replayed on a Pointe-à-Pitre to Ushant TSS feeder lane: 52 weekly departures on ERA5 winds, waves and GLORYS currents, scored against an honest rhumb baseline. About 2.5% median fuel saved at held speed and 6.0% with just-in-time arrival, of which roughly half is geometry. Every passage and its weather is listed, with the hindsight-ceiling caveat stated.
A laden passage southwest of Japan with Typhoon Dolphin across the track. WindMar's isochrone optimiser routes the ship around the storm core at the same engine order: 4.9% less fuel over a longer track, but the point is safety: peak roll cut from 33.9° to 17.0° and significant wave from 7.4 m to 5.8 m, with the seakeeping module flagging the synchronous-roll legs.
A factorial fuel-gain study on a Gibraltar-New York crossing for an MR tanker, plus a worked seasonal expectation: least-fuel routing over coupled wind, sea-state and current fields. The gain is about 2% of voyage fuel at the physics ceiling, concentrated in heavy weather; the great circle remains a strong baseline.
A geometry / seasonal-β / forecast-α decomposition of optimizer fuel-ROI across ocean regimes, under ERA5+GLORYS replay against a pre-registered protocol. The forecast-dependent component a real-time service adds is under 1% of fuel: and that is a hindcast upper bound.
Independent cross-validation of WindMar's routing engines on five diverse routes against a commercial weather-routing reference: comparing route geometry, weather conditions encountered, and ETA. No fuel comparison.
Complete guide to building a self-hosted AI automation pipeline using Docker, n8n workflows, Ollama for local LLMs, and Telegram for notifications.
How to Shift Projects in a Rapidly Evolving Environment - exploring the benefits and practical considerations of running large language models locally.
Analysis of the arbitral award in the Gulf of Piran maritime dispute between Slovenia and Croatia.
Comprehensive guide to tanker operations: oil tankers, chemical carriers, and gas carriers.
Design your own AI teaching assistant.
Interactive AI assistant for learning Slovenian language.
Interactive AI assistant for learning Italian language.
Building Dual-Agents Chatbots with Python: Harnessing AI for Targeted Knowledge Acquisition and Automated Learning.
AI-driven analysis of future global trends and policy implications.
Multi-agent AI conversation exploring climate policy challenges.
Architecture and design patterns for multi-agent AI systems.
Can ChatGPT help legal practitioners?
The difficult exercise of balancing risk and reward.
Blending Momentum and Mean-Reversion scorings.
From Schrodinger equation to technical trading.
Returning 18% CAGR since 2020 in an underlying bearish market.
Achieving a 7% Return with minimal market exposure.
How to use Bollinger Bands and RSI to hedge physical asset.
A presentation of the QuantCoder project.
Integrated Extraction and Coding of Alpha Research.
Building an AI trading assistant with MyShell.
Interactive AI trading assistant application.
Essential Data Gathering for Quantitative Workflows.
A step-by-step guide using IB_insync.
Optimizing Python and QuantConnect Development with AI.
How a Simple Script Can Preemptively Guide Your Trading Decisions.
A Study on ROA and ROE's Efficiency in Low-Valuation Universe with Quantconnect.
Beyond a bootstrap Calculation of Value-at-Risk.
Developing securities universe classes in Python.
Universes, Alpha signals and Portfolio models using ib_insync.
An extension of previously introduced AI-driven stocks selection methodology, focusing on actionable Python code.
Fundamental screening techniques for value investing.
Machine learning approaches to stock clustering and portfolio construction.
Time series analysis for intraday market prediction.
Testing ARIMA models for market direction forecasting.
Comparing traditional and deep learning approaches to time series forecasting.
