Hu Bowen

• GPA: 3.98/4.00
• Diploma with Merit (7 module prizes; placed first for 3 modules)
• DSTA Polytechnic Digital Scholar
• Vice-President, NullSec (Cybersecurity Interest Group)

• Maintained an internal Large Language Model (LLM) safety evaluation tool
• Researched and benchmarked adversarial attacks on systems using modern Natural Language Processing (NLP) techniques

Implemented Advent of Code solutions on an FPGA with Jane Street's DSL

• Wrote logic similar to the Game of Life by using shift registers and line buffers to represent a convolution window
• Achieved a 100x speedup relative to my naive solution in OCaml

Bare metal operating system from scratch

• Implemented a bootloader in x86 assembly to load a C kernel in 32 bit mode
• Wrote a basic command parser and a text-based Snake game directly in the kernel

AI competition testing across ASR, CV, NLP and RL

• Placed 1^st in the novice category during qualifiers
• Fine-tuned multiple RT-DETR models for CV and routed images depending on model confidence
• Implemented Proximal Policy Optimization (PPO) with Prioritized Fictitious Self-Play (PFSP) for RL
• Ported the entire RL environment onto the GPU by rewriting it in JAX, achieving a 1000x speedup in raw FPS

Only Polytechnic student to attain Gold in NOAI 2026

• Answered 20 in-depth MCQs on various technical aspects of AI
• Wrote a gradient boosting regressor in Python; debugged and completed a PyTorch U-NET implementation
• Implemented multi-head scaled dot product attention with PyTorch

National cybersecurity competition held over 3 days

• Configured Cisco firewalls to isolate a cloud environment from the internet
• Only competitor to successfully exploit a vulnerable web application, placing us in first place for that component