• Final-year Software Engineering undergraduate at Sichuan University.
• Research Intern at the Machine Intelligence Lab (MIL), working on multimodal learning for medical imaging and clinical data.
• Experienced in deep learning, computer vision, generative models, and data-intensive software.
• Seeking graduate research opportunities in AI and Computer Science.

Sichuan University · Chengdu, China
B.Eng. in Software Engineering · Sep 2023 – Jul 2027 (Expected)

• GPA: 3.41 / 4.00
• IELTS Academic: 7.5

Research Intern · Machine Intelligence Lab (MIL), Sichuan University
Sep 2026 – Present
Undergraduate thesis: Multimodal Source-of-Gain Analysis for chest X-ray findings

• Building a missingness-aware classifier that fuses chest X-rays with ECGs and blood tests on MIMIC data.
• Designing controls to check whether accuracy gains come from the extra modalities.
• Completed a feasibility pilot; now finalizing the protocol before the full experiments.

FaRe-CXR Explainable Chest X-ray Classification
May 2026 – Present
International research collaboration · Türkiye

• Ran multi-label classification experiments with DenseNet-121 and multi-scale feature extraction on NIH ChestX-ray14.
• Evaluated five findings across 27 runs (nine configurations × three random seeds).
• Reached a validation macro-AUROC of ~0.83–0.84 and studied explanation faithfulness.

Big Data Analytics Intern · Chengdu Suncaper Data Technology Co. Ltd.
Feb 2026 – Apr 2026
Project: YelpInsight · Repo · Demo

• Processed ~6.7 million Yelp reviews with Apache Spark, PySpark, and Hive.
• Integrated NOAA and Census datasets to derive 50+ analytical metrics.
• Built interactive analytics dashboards with React, TypeScript, and FastAPI.
• Contributed to a schema-aware natural-language-to-HiveQL application powered by DeepSeek APIs.

FROST - Fuzzy-Rough Structure Preservation
Generative Models · Image-to-Image Translation · Granular Computing

• Led a three-person project integrating fuzzy-rough structure masks into CycleGAN-Turbo training.
• Compared four mask-generation methods across two datasets and both translation directions.
• Fuzzy masks achieved the lowest FID in all four settings, improving it by 11.2% on Horse→Zebra and 9.6% on Day→Night.

Closed-Form Concept Erasure via Double Projections
Diffusion Models · Concept Erasure · Model Editing

• Reproduced a closed-form concept-erasure method on Stable Diffusion 1.4.
• Cut target-concept classification accuracy from 76% to 2%, with only a 2.8% average drop on other classes.

OSD-Insight
Object Detection · Open-Set Recognition · Uncertainty Estimation

• Led a three-person project on unknown-object rejection with a frozen Faster R-CNN ResNet-50 FPN detector.
• Evaluated confidence, energy-based, and semantic-uncertainty scoring on Open Images V7.
• The energy-based baseline reduced A-OSE from 190 to 93; the adaptive approach raised AP from 0.485 to 0.505.

Clairo
Android Development · AI Voice Assistant · On-Device Inference

• Led a five-person team building an Android AI voice assistant.
• Integrated Flutter, Kotlin, ONNX Runtime, SenseVoice, and LLM APIs, with support for user-provided API keys.

• 🥉 Third Prize - vivo AIGC Track, China University Computer Competition, VIVO x Nankai Univeristy (2026)
• 🏆 Grand Prize - National University Governance Case Analysis Competition, Fudan University (2026)
• 📜 Excellent Student Certificate - HIT Global Summer School (2026)
• 🎖️ Finalist - Delulu Hack, PayNet Malaysia (2025)
• 🎓 Belt and Road Scholarship - Sichuan University (2023, 2024, 2025)