Industry Internships

Shanghai Qianshiwei Digital Technology (A High-Frequency Quant Trading Startup)

Quantitative Model Research Intern
May 2026 - Jul 2026, Shanghai, China

Futures price sequence prediction using an Encoder-Decoder architecture:

1. Filtered high signal-to-noise intervals from 10 million 500-ms SHFE silver observations over 20 months through microstructure analysis.
2. Designed a tokenizer and vocabulary construction method for market data, mapping continuous values into discrete tokens.
3. Built an Encoder-Decoder (T5) architecture to learn high-dimensional representations of market data and predict future price sequences.
4. Performed representation interpretability analysis on a self-modified T5 architecture, tracing how its attention, normalization, feed-forward, and FiLM modules transformed input sequence representations.
5. Applied PCA, KNN, Cohen’s d, correlation analysis, cosine similarity, and heatmap visualizations to assess whether K/V channels and layer-wise hidden states captured intended price-volume sequential structures.
6. Designed and built a large-scale hyperparameter search system that orchestrated experiment planning, execution, evaluation, analysis, memory management, and fault recovery into an Agent Loop and Harness spanning 59 hyperparameters.

Binance

Data Scientist (AI/LLM) Intern, AI-Data Services Department
Sept 2025 - Feb 2026, Remote/Singapore

Built data pipelines for an LLM-based customer-service system comprising an intent LLM and an answer LLM:

1. Developed an automated intent-classification pipeline for SFT data, iteratively deriving category-boundary rules across 4 top-level and 68 sub-categories. Achieved 95% precision, 90% accuracy, and 90% recall.
2. Cleaned and decomposed rule-based bot replies into independent cases, then performed embedding-based semantic deduplication to produce a million-scale Q&A dataset.
3. Built a query-clarification dataset for RAG, improving the system’s handling of ambiguous queries.
4. Built an SOP knowledge base and a pipeline for mapping user queries to internal customer-service procedures.

ByteDance

LLM Algorithm (AI4Code) Intern, Trae-MarsCode Team
May 2025 - Aug 2025, Shenzhen, China

Optimized the Code Completion Model in Trae through data engineering, reinforcement learning, and DPO:

1. Contributed to the design and implementation of Context Engineering, merging Trae user interaction data with model completion code data.
2. Analyzed rejected code suggestions, refined the labeling system through selective annotation, and clarified optimization targets.
3. Built an automated data-labeling pipeline with LLM-as-a-Judge, Auto-Prompt, and Majority Vote, improving its accuracy to 82% and preparing data for RL/DPO training.
4. Introduced a critic LLM to evaluate bad-case annotations and built a knowledge base for iteratively improving judge LLM prompts.

Vitaly Medical Inc.

Machine Learning Intern
Oct 2024 - Feb 2025, Berkeley, USA

Utilized the open-source wrist-worn accelerometer dataset WEDA-FALL to design and implement a fall detection system:

1. Applied data preprocessing techniques including filtering, segmentation, and feature extraction.
2. Designed and trained an ANN model for fall detection, obtaining an accuracy of 98.6%, and converted it to TensorFlow Lite for mobile deployment.
3. Enhanced the system by implementing a post-fall inactivity recognition feature.

Baidu

Speech Algorithm Intern, Speech Technology Department
Jun 2024 - Aug 2024, Beijing, China

Contributed to a speech LLM for intelligent cockpit applications:

1. Performed slot normalization to convert spoken numerical expressions in English into real numbers, percentages, and phone numbers.
2. Parsed parameter conditions obtained through intent recognition and converted them into specific function calls.
3. Used Llama3-8B to generate chat data, including 20,000 semantically neutral samples, 20,000 single-turn dialogues, and 5,000 multi-turn dialogues.
4. Recursively validated and corrected JSON data from LLM responses using Python and the Llama3-8B interface.
5. Cleaned a raw dataset to obtain millions of query-intent pairs through string parsing, similarity filtering, and duplicate removal.
6. Independently developed the human-feedback features of a Multi-Turn Task-Oriented Dialogue Model for RLHF data labeling, reconstructed its NLU module, and implemented Streamlit interactions for colleagues.

BNP Paribas (China)

Corporate Banking Intern, CIB Department
Jun 2023 - Aug 2023, Shanghai, China

Main jobs:

1. Collected quarterly and annual financial data and analyzed their fluctuations.
2. Analyzed and evaluated corporate clients’ ESG reports based on BNP Paribas standards.
3. Automated batch searching with Python to support team members in industry research.

Shenzhen NINES Internet Co., Ltd. (Campus Start-up)

Front-end Engineer Intern, Dev Department
Sep 2022 - Dec 2022, Shenzhen, China

Main jobs: Collaborated to Build an Online Information Platform for the Campus.

• Received demands from the product department, including UI design and interaction logic;
• Based on the Uni-app (HBuilder X), collaborated to develop (using Vue.js) the pages of a WeChat mini-program.