Profile
CS Master's student at USC with a strong background in NLP, Deep Learning, and AI. Experienced in developing language models, NER systems, and multimodal retrieval pipelines. Passionate about Trustworthy AI and Generative Models.
Education
Master of Science, Computer Science
Aug 2024 - Present at University of Southern California, Los Angeles, CARelevant Coursework: Deep Learning, Algorithms, Artificial Intelligence, NLP, Databases, Robotics.
and Bachelor of Arts, Economics
Sept 2020 - June 2024 at University of California, Santa Cruz
Highest Honors in Computer Science; Dean's Award; University Honors, Cum laude.
Relevant Coursework: Data Structure and Algorithms, Intro to AI, Applied ML, Advanced NLP.
Work Experience
Small Language Model Algorithm Intern
June 2025 - Aug 2025 at Rokid Glasses, Hangzhou, China
- Designed and delivered an end-to-end 30-language TTS generation and evaluation pipeline.
- Combined FastText + LaBSE for language ID and integrated Coqui, Kokoro, and MMS TTS engines.
- Implemented GPU batch concurrency and thread-safe model loading for optimization.
- Built an automated quality evaluation system using Whisper-turbo, LaBSE, and MuSR.
- Constructed a multi-dimensional scoring suite (BLEU, chrF++, BERTScore, COMET) to guide model updates.
Research Experience
Named Entity Recognition Models (Graduate Research)
Feb 2025 - Mar 2025 at USC, Los Angeles, CA
- Developed a BiLSTM NER model achieving 94.60% accuracy and 73.70% F1-score on CoNLL-2003.
- Enhanced architecture with 100-dim GloVe embeddings, boosting F1-score to 82.18%.
- Implemented a BiLSTM + CNN hybrid model, achieving 96.64% accuracy and 85.95% F1-score.
Attention Bias Mitigation in LLMs (Graduate Research)
Sept 2024 - Dec 2024 at USC, Los Angeles, CA
- Utilized fine-tuning and data augmentation on Qwen 2.5 0.5B model to mitigate attention biases.
- Achieved 77.07% F1 Score on SQUAD and 62.04% on HotpotQA.
- Implemented attention reweighting strategies inspired by PASTA method.
Real Vision Question and Answer Dataset (Undergraduate Research)
Sept 2023 - Feb 2024 at UCSC, Santa Cruz, CA
- Implemented a multimodal retrieval system leveraging GPT-4 and LangChain for agriculture and healthcare queries.
- Evaluated hallucinations in LLM outputs and highlighted areas for improvement in information accuracy.
Skills
- Python
- PyTorch
- Tensorflow
- Transformer
- NLP
- GenAI
- Docker
- AWS/GCP
- Git
- Pandas/Numpy
- R/Stata
- Computer Vision