About
AI Application Development
I'm Yuanhao Feng, pursuing a Master of IT at the University of Melbourne with a focus on artificial intelligence. My undergraduate major was software engineering.
In AI application development, what takes the most time is rarely the model itself, but the engineering that makes the model run reliably within a complete system. Whether the evaluation framework is sound determines if optimization is driven by data or intuition; how cleanly model and code responsibilities are separated determines whether edge cases are handled reliably. From model evaluation and selection, through agent orchestration and retrieval-augmented generation, to API design, deployment, and cost control, I'm used to owning things end to end.
My independent projects span a fairly wide range: AI applications, pure engineering, game-adjacent tools, and more. Most start as solutions to a small need of my own, but once I begin building, the focus naturally extends beyond just making it work: whether a new user can get started smoothly, how to handle failures gracefully, how costs hold up over the long term. That arc, from a small personal tool to a product accountable to real users, is a process I find genuinely compelling.
This site is mine too, from the server all the way to the page. A selection of what I have built lives in Projects.
Experience
AI Application Developer, Intern
COSCO Shipping Technology · Shanghai
Dec 2025 – Feb 2026
Led the container-number recognition technology evaluation, benchmarking six vision-language models and driving the team's shift from traditional OCR to a multimodal approach, raising the success rate from 37.5% to 98.3%. Also built a structured field-extraction agent for shipping emails with an automated evaluation pipeline, achieving 98.8% field-level accuracy.
Software Developer, Intern
COSCO Shipping Technology · Shanghai
Nov 2024 – Jan 2025
Independently developed a bill-of-lading PDF extraction pipeline end-to-end, designing a two-stage architecture that pairs OCR for spatial detection with an LLM for semantic understanding, and shipped it as a REST API. Also designed a geometric correction scheme for scanned documents that handles orientation detection and coordinate transformation in a single OCR pass, eliminating the overhead of a second recognition run.
Education
The University of Melbourne
Master of Information Technology · Artificial Intelligence
Feb 2025 – Dec 2026
Nanjing University of Information Science & Technology
B.Eng. in Software Engineering
Sep 2020 – Jun 2024