Profile
M.Eng. Candidate in New Generation Electronic Information Technology (including Quantum Technology, etc.), Hainan University | CPC Member | National Scholarship Winner
Algorithm Research
M.Eng. candidate in Electronic Information at Hainan University (Class of 2027), researching computer vision and open-world individual recognition. First author of two EI international conference papers; I independently completed the full research loop from algorithm design and model training/tuning through ablation validation. My research proposed a quality-aware mechanism and the Compact256 compact projection layer — the former turns imaging-quality fluctuation into a learnable signal, the latter compresses features to 1/5 while holding accuracy loss within 2% — so I bring a complete research methodology: spot the problem, abstract it into an algorithmic problem, design a solution, validate it experimentally.
Engineering Delivery
Algorithms do not stop at the paper: I completed PyTorch → ONNX conversion and edge-deployment adaptation, built an online inference service on FastAPI + MySQL + Nginx, designed five runtime mechanisms for on-site stability problems, and brought the false-acceptance rate in open scenarios down to 6.91%. Six months on site delivering and iterating the system taught me the engineering distance between “model accuracy” and “system availability”.
Edge AI Deployment
I completed lightweight compression of a deep-learning model (features compressed to 1/5 with accuracy loss <2%) and ONNX edge-inference adaptation, developed an online inference service with a Python/C mixed approach, and designed five runtime stability mechanisms for real on-site constraints; the system reached the engineering freeze standard and has run stably in production for more than six months. Working on site taught me the real constraints of edge systems — compute, memory, bandwidth, thermals and network, each of which erupts where the lab cannot see it.
Debugging & Root-Cause Analysis
I follow a systematic debugging routine — reproduce the symptom, attribute it to a module, verify minimally, then verify the fix: I located a USB bandwidth bottleneck and multi-thread contention in the acquisition-terminal project, and tracked down track breaks and frame-quality problems on site in the recognition system. I have C/C++ and Linux fundamentals, am familiar with multi-threaded programming and synchronization, can read hardware schematics, and have hardware-software co-debugging experience.
Learning Ability & Potential
GPA 3.6/4.0 in graduate school; as an undergraduate I received the National Scholarship (awarded to roughly 0.2‰ of students nationwide) and the provincial Outstanding Graduate award; and I won a national 3rd prize in the “Datang Cup” National Collegiate Mobile Communication 5G Challenge. From vision-language model fine-tuning to edge deployment and full-stack service development, I keep extending beyond my algorithm comfort zone, with learning transfer validated across multiple directions.
Overall Capability
I served as a club president and as head of a fire-safety volunteer centre, so my organizing, coordination and cross-department communication skills are battle-tested; I have long written technical blogs and express technical ideas clearly. I believe the value of an algorithm lies in solving real problems, and I want to turn academic accumulation into product competitiveness in industry.