At EY, I got to see what it really means to apply AI to real-world business problems.

Instead of working on AI in isolation, I started from the ground level — understanding how logistics teams operate day-to-day, where delays come from, and why certain decisions (like routing or workforce allocation) are harder than they seem. Once I understood the workflows, I mapped areas where AI could reduce manual effort and improve efficiency across transportation, warehousing, and distribution.

I then focused on three practical use cases — routing optimisation, demand forecasting, and workforce planning — and helped estimate the potential business impact, which came out to roughly 5–10% cost and productivity improvements depending on adoption.

One of the most valuable parts of the internship was working with messy, real operational data (10k+ records across sources). It taught me that the hardest part of “AI” is often not the model — it’s defining the right problem, cleaning the data, and making the output usable for decision-makers.

By the end of the internship, I walked away with a clearer understanding of how strategy and technology combine — and how meaningful outcomes come from a mix of structured thinking, collaboration, and communication.

Work Experience

Ernst & Young

Consulting Intern, AI in Logistics

Hokkaido University

AI Researcher

At Hokkaido University, I worked on a research project that felt deeply meaningful: using AI to make disaster reporting faster and more reliable.

In a world where minutes matter, our goal was to build a system that could detect disasters across the globe within a few hours, identify where they occurred, and estimate how severe they were. We combined NLP, clustering, and location verification to transform unstructured news articles into structured disaster intelligence.

My work focused on NLP. Using BERT and spaCy, I helped build models that could analyse disaster-related news with around 90% accuracy. We then organised more than 1,000 disaster events using K-means clustering and validated over 1,200 locations with strong precision. I also developed a custom disaster severity scale based on deaths, injuries, and damage — enabling better prioritisation and decision-making.

The results were exciting, but the most memorable part was the collaboration: working with researchers from different cultures and learning how global teams build meaningful systems together. This project is now being formalised into a research paper, and it reinforced something I strongly believe — technology is most powerful when it is built for real human impact.