Curriculum Vitae

Jiawei
Lu

Building at the intersection of data and AI · MS Statistics (Machine Learning Track) @ Columbia University · Incoming Summer 2026 SWE Intern @ xHealth · Ex-Stepfun, Red Note

New York, USA fayette0302@gmail.com LinkedIn GitHub
MS Statistics student (Machine Learning track) at Columbia University (2025–2027), with hands-on experience in LLM data pipelines, multi-model inference systems, and data-driven business analytics. Previously interned at Stepfun (阶跃星辰), Red Note (小红书), and Continental. Joining xHealth Group in Dallas as a Summer 2026 Engineering Intern, leading the build of a conversational AI health agent.

Internships

Jun 2026 – Aug 2026

Dallas, TX

Incoming

xHealth Group

Summer Engineering Intern

A 10-week, build-first internship: own a conversational AI health product end-to-end — from concept through Demo Day — while collaborating on an internal AI sales & marketing platform.

  • Leading development of a multi-session conversational AI agent that maintains persistent patient context for symptom tracking and medication adherence workflows.
  • Designing escalation logic to surface high-risk patient signals and generating structured summaries for clinician review — outputs must be clinically usable, reliable, and auditable.
  • Building LLM orchestration, context/session management, and backend services for structured interaction storage with an API-ready architecture.
  • Collaborating on a three-person team building an internal AI platform for automated outbound messaging, AI-generated marketing content, lead segmentation, and multi-channel campaign orchestration.
LLM Orchestration Conversational AI Context Management Backend Healthcare AI

Jun 2024 – Aug 2025

Shanghai, China

Shanghai Stepfun Co.

LLM Data Engineering Intern

Training a large language model starts long before any gradient update — it starts with data. At Stepfun, I worked on the data pipeline side of in-house LLM development, focusing on Chinese conversational data sourced from crosstalk (相声) recordings.

  • Designed a three-stage filtering pipeline (rule-based regex → LLM-as-a-Judge binary classification → manual spot-check) to process 10K+ raw dialogues into ~6K gold-standard training samples.
  • Developed few-shot examples to guide LLM-assisted rewriting of low-density dialogue segments, improving training signal quality.
  • Built a multi-model routing pipeline (GPT / Claude / Qwen) with async API execution (~5× speedup) and local parallelization for benchmarking (~10× speedup).
  • Ran prompt placement ablation studies across models, finding that optimal System-role vs. User-role injection varied by model — insights that directly shaped pipeline design.
  • The curated dataset contributed to a 6–7 point gain on internal reasoning/language benchmarks, clearing the bar for consumer deployment.
Python Prompt Engineering LLM Data Pipeline API Integration

Feb 2024 – May 2024

Shanghai, China

Red Note

Data Analyst Intern, FMCG Department

Red Note's sales team needed to show brand clients — mostly in the health & wellness space — whether their ad spend was actually working, and how they stacked up against competitors. I built the data infrastructure and analysis that made those conversations possible.

  • Automated data retrieval from Red Note's internal analytics platform via Selenium, cutting data pulling time from half a day to under 30 minutes.
  • Built Excel-based multi-dimensional analyses tracking return-search rate and conversion rate across product lines, time periods, and placement types for 10+ contact lens brands.
  • Produced anonymized competitive benchmarking reports, giving clients a view of peer performance without revealing brand identities.
  • Delivered strategic recommendations to sales and category managers; 4 key accounts adopted nearly all suggestions, shaping subsequent budget decisions.
  • Assisted in post-campaign wrap-up by preparing data components of recap decks, summarizing performance trends for internal and client-facing review.
Python (Selenium) Excel SQL Data Analysis Data Visualization

Jun 2023 – Sep 2023

Shanghai, China

Continental Tires

Operations Analyst Intern

On the distribution side of a global automotive brand, keeping dealers accountable at scale is a logistical challenge. I helped make that process faster and more systematic.

  • Analyzed dealer-level sales data to identify underperforming stores and flag those requiring formal intervention.
  • Automated warning letter distribution via mail merge, reducing turnaround from ~2 days to a single afternoon for ~10–15 letters per peak cycle.
Excel Data Analysis

Projects

Dec 2025

Columbia University

NLP Course Project

NL to Performant SQL (NL2SQL)

Group Member

Can we teach a small model to generate not just correct SQL, but efficient SQL? We fine-tuned Qwen3-4B using a two-stage alignment pipeline on the BIRD benchmark.

  • Fine-tuned Qwen3-4B-Instruct for natural language to SQL generation using Direct Preference Optimization (DPO) on curated efficient vs. inefficient SQL pairs, followed by execution-aware reinforcement learning (GRPO-style) with latency-based rewards.
  • Achieved a 7% improvement in execution accuracy and Relative Valid Efficiency Score (R-VES) over baseline on the BIRD Mini-Dev set.
  • Conducted qualitative error analysis on query structure and optimization patterns.
PyTorch HuggingFace DPO GRPO NLP SQL
View Report →

Feb 2026 – Present

Columbia University

Stats ML Course Project

Crime Data Classification Analysis

Group Leader

A systematic comparison of feature selection and regression methods on the UCI Communities and Crime dataset — plus an interactive dashboard to make the results explorable.

  • Analyzed the UCI Communities and Crime dataset using multiple classification and regression approaches including Logistic Regression (One-vs-Rest), Multinomial Regression, Naive Bayes, Linear Discriminant Analysis.
  • Performed systematic model comparison with cross-validation, evaluating predictive performance, interpretability, and robustness to multicollinearity across methods.
  • Built an interactive data visualization dashboard with vanilla JavaScript, featuring model comparison charts, feature selection consensus matrices, coefficient heatmaps, and hyperparameter grid search visualizations.
Python Scikit-learn JavaScript Data Visualization Machine Learning
Interactive Dashboard →

Volunteer & Leadership

Dec 2022 – Jan 2025

Shanghai, China

SJTU SEIEE Student Volunteer Association

Vice President & Head of Publicity

Over two years, I grew from taking photos at events to co-running a 130-person organization — learning that volunteer coordination is as much about logistics and communication as it is about goodwill.

  • Progressed from Publicity Officer → Head of Publicity → Vice President over three consecutive years.
  • Co-led ~2 events per month, owning end-to-end coordination of headcount, materials, schedules, and external stakeholder communication.
  • Managed a 25-person publicity team, overseeing content production on WeChat Official Account and other platforms.
  • Led the full publicity cycle for Shanghai Marathon volunteer program — from recruitment to post-event recap — one of the organization's most visible annual commitments.