UC San Diego ECE M.S. | Machine Learning & Data Science | LLM Systems, Trustworthy ML, Data-centric AI
Based in San Diego. I am looking for Fall 2026 research opportunities with UC San Diego labs working on reliable AI systems, LLM agents, data-centric machine learning, and ML for scientific or high-stakes decision support.
I build machine learning systems where model outputs need to be grounded in source data, deterministic computation, and measurable evaluation. My background combines UCSD graduate coursework in machine learning for physical applications, statistical learning, big network data, and computational evolutionary biology with applied projects in LLM agents, RAG, entity extraction, risk analytics, and high-throughput ML services.
Current research interests:
- Reliable LLM and agent systems: tool use, retrieval, evaluation, hallucination control, and auditable decision pipelines.
- Trustworthy and responsible ML: confidence calibration, human-in-the-loop review, interpretability, fairness monitoring, and robust evaluation.
- Data-centric AI and ML systems: document/entity extraction, weak supervision, data pipelines, experiment tracking, and scalable model serving.
- ML for physical and high-stakes systems: time-series anomaly detection, portfolio risk analytics, energy dispatch, and scientific machine learning.
| Project | Focus | Why it matters |
|---|---|---|
| MindMarket / PersonalFinancialRiskManagement | LLM-grounded risk analytics | Combines deterministic VaR/CVaR, factor beta, and stress-scenario computation with an AI copilot that is grounded on computed evidence rather than free-form numerical generation. Relevant to reliable AI for high-stakes decision support. |
| Multimodal RAG System | Retrieval, embeddings, and evaluation | Integrates CLIP vision embeddings, text embeddings, rank-fusion reranking, FastAPI serving, and Recall@5 evaluation. Relevant to LLM reliability, retrieval quality, and grounded generation. |
| EDAgent | Tool-using optimization agent | Uses a ReAct-style agent with CVXPY tools for economic dispatch, separating numerical optimization from language generation. Relevant to AI agents for scientific/engineering workflows. |
| Kaggle LMSYS Chatbot Arena | LLM preference modeling | Kaggle Silver Medal project involving large-scale model inference, preference prediction, and performance optimization. Relevant to LLM evaluation and ML systems. |
| Amazon Review Deception Detection | NLP classification | Applies text classification to deceptive-review detection, connecting NLP, trust, and high-signal evaluation. Relevant to data-centric NLP and trustworthy ML. |
| Role | Research-relevant work |
|---|---|
| Machine Learning Engineer Intern, Permitfolio | Built an auditable LLM-assisted asset-classification service for RegTech compliance with human-in-the-loop review, confidence calibration, review flags, model-version tracking, and reproducible tests. |
| Machine Learning Research Assistant, Penn State | Built an NLP entity-extraction pipeline with spaCy, Transformers/BERT fine-tuning, Kafka streaming, MLflow experiment tracking, Optuna tuning, and evaluation metrics. |
| Machine Learning Engineer Intern, Allianz Insurance | Built fraud detection and model-serving pipelines with privacy-preserving feature engineering, fairness monitoring, Redis-backed serving, and latency optimization. |
| Area | Tools |
|---|---|
| ML / AI | PyTorch, Transformers, BERT, LLMs, RAG, CLIP, embeddings, scikit-learn, MLflow, Optuna |
| Data / Systems | Python, SQL, Kafka, FastAPI, Flask, PostgreSQL, Redis, PySpark, Docker, Linux |
| Evaluation | Recall@K, F1, precision/false-positive analysis, confidence calibration, fairness metrics, pytest, JMeter |
| Engineering | TypeScript, Next.js, GitHub Actions, AWS EC2/Lambda/S3, CI/CD, observability |
UC San Diego ECE M.S. coursework includes Machine Learning for Physical Applications, Computational Evolutionary Biology, Machine Learning for Music, Statistical Learning, Big Network Data, Linear Algebra and Applications, and Smart Grids.
I am open to UC San Diego research collaborations starting Fall 2026, especially projects that can lead to a workshop, conference, or systems paper.
- LinkedIn: linkedin.com/in/zhengdong17
- GitHub: github.com/zhengbrody
Copyright © 2026 Zheng Dong. This profile README is personal portfolio content and is not licensed for reuse, redistribution, or impersonation.