Hi, I'm Jiaran (Jay) đź‘‹
Data-Driven Decision Making
Turning messy, real-world data into clear stories that drive strategic decisions and transform business outcomes.
About Me
Hi, I'm Jiaran — I'm an Applied AI Data Scientist at Ideate Capital, a startup where I build AI-powered pipelines that turn messy, real-world data into insights that drive business decisions. I hold a Master's in Data Science from NYU and bring hands-on experience with LLMs and deep learning alongside classical machine learning and statistics. I'm passionate about uncovering patterns, telling stories with code, and building tools that connect data to real-world decisions. Whether designing predictive models, fine-tuning language models, or crafting intuitive interfaces, I love solving meaningful problems at the intersection of statistics, technology, and human experience.
Outside of work, you'll find me exploring new cities and cultures, water diving, — or getting unreasonably excited about lifting a heavy weight.
Skills
Programming
Python (NumPy, Pandas, Scikit-learn, LightGBM), SQL, R
Statistics
Regression Analysis, Hypothesis Testing, Time Series Forecasting, A/B Testing
Machine Learning
Classification, Regression, Clustering, PCA, Ensemble Methods, Deep Learning (CNN, RNN, PyTorch, TensorFlow), Fine-Tuning (LoRA/PEFT)
GenAI
Transformers, Hugging Face, Prompt Engineering, LLM-as-Judge, RAG, LangChain
Other Tools
Hadoop, Spark, AWS, Snowflake, Tableau, Power BI, JIRA, Git
Professional Experience
Applied AI Data Scientist
Ideate Capital
New York, NY
February 2026 – Present
- Translated a consumer brand's declining sales question into a data pipeline mining alternative data (Reddit, TikTok, Instagram) — tracking sentiment trends, conversation volume, and topic themes across 2M+ posts to inform go-forward strategy.
- Built a scalable AI-powered enrichment pipeline, adding 50+ fields (revenue, ownership, leadership) across 300,000+ companies, managed in Snowflake with full metadata and version tracking.
- Built an iterative QA validation loop combining SQL-based checks and LLM-as-judge to catch errors, trace them to their enrichment step, and refine the pipeline — raising accuracy from under 80% to 97%.
Data Scientist Intern – Product & Growth
ETH Tech
New York, NY
July 2025 – September 2025
- Identified a 70% cart abandonment rate across 1M sessions as the key revenue bottleneck in an e-commerce funnel.
- Built and A/B tested a real-time notification feature — targeting users with abandoned carts and customizing content to their specific items — driving a 15% lift in conversion vs. generic notifications.
- Prototyped a conversational shopping assistant using context injection (live inventory data) and structured output to clarify customer intent and generate product recommendations.
Group Leader
Probability of Default Modeling – Applied ML in Finance Project
New York, NY
October 2024 – December 2024
- Built an end-to-end probability of default prediction system using a 1M-row SME loan dataset: engineered 14 financial ratios, applied rule-based balance sheet imputation, quantile binning, and one-hot encoding.
- Developed a grouped ensemble of 9 LightGBM models, trained in a walk-forward fashion with false negative penalties and weighted predictions to account for class imbalance and credit risk.
- Validated model performance on time-ordered out-of-sample test sets; used SHAP plots to interpret key features; achieved 0.875 AUC, outperforming the unsegmented logistic regression baseline (0.78 AUC).
Data Scientist Intern
Guotai Junan Securities
Shanghai, China
June 2024 – August 2024
- Built a bond yield prediction pipeline using decision tree models, achieving a 0.73 F1-score on test pricing data.
- Developed a stepwise regression model to estimate fund duration, reducing duration volatility by 25%.
- Created a real-time repo rate monitoring tool by integrating external data into Excel, enabling downstream modeling and improving fixed income strategy responsiveness.
Quantitative Research Intern
ZADS Fund
Shenzhen, China
June 2023 - August 2023
- Engineered 16 high-frequency alpha factors using market microstructure data (Level-2/order book), inspired by academic literature and proprietary research.
- For each factor, calculated signal values, evaluated predictive power, and backtested performance using historical intraday data.
- Built a machine learning pipeline integrating 10 low-correlation factors to predict T+1 stock returns, achieving an AUC of 0.62 on out-of-sample data.
Data Scientist Intern
China Everbright Bank
Beijing, China
June 2022 – August 2022
- Processed 200K+ customer records from 8 relational tables and reduced dimensionality from 700+ to 50 features using WOE encoding and Information Value (IV) selection.
- Built and compared binary classification models (logistic regression, decision tree, random forest) to predict customer asset change; selected logistic regression for its interpretability and high AUC (0.85).
- Delivered model insights through dashboards, contributing to a 20% lift in conversion and 12% reduction in acquisition cost.
Featured Projects
Fine-Tuned Language Model for Social Media Sentiment & Topic Classification
Fine-tuned RoBERTa-large with PEFT (LoRA) for sentiment classification on 1M+ labeled social media posts/comments, enabling parameter-efficient training with only ~0.3% trainable parameters.
Probability of Default Modeling
Built a probability of default prediction pipeline on 1M+ loan records—cleaned data, engineered 14 financial ratios as features and applied machine learning to model credit risk. Applied SHAP for model interpretability, quantifying each feature's impact on default predictions.
Bond Yield Prediction
Built a bond yield prediction pipeline using decision tree models for Guotai Junan Securities, and developed stepwise regression models for fund duration estimation.
Customer Asset Prediction
Processed 200K+ customer records from 8 relational tables; used WOE and IV to reduce 700+ features to 50, building binary classification models to predict asset change.
Quantitative Investment
Engineered 16 high-frequency alpha factors using market microstructure data (Level-2/order book), inspired by academic literature and proprietary research. Built a machine learning pipeline integrating 10 low-correlation factors to predict T+1 stock returns.
Statistical Analysis of Movie Ratings
This data analysis project explores movie ratings from over 1,000 participants across 400 films. I used statistical testing to investigate patterns in preferences—like how gender or birth order might influence opinions on Shrek or The Lion King. I also built regression models using Ridge, LASSO, and Elastic Net to predict ratings based on viewer demographics.
Beyond the Code
🏋️ Weightlifting
Passionate about strength training and pushing physical limits. Love the discipline and progress that comes with consistent training. Whether it's powerlifting, progressive overload, or just getting unreasonably excited about lifting heavy weights.
During my undergraduate years, I competed in a 15-event all-around fitness competition — including powerlifting's big three (squat, bench press, deadlift), sprints, standing long jump, endurance runs, and more. I won 9 individual events and took home the overall championship title.
đźš— Solo Road Trip
Cross-country adventure, covering 4000 miles across 14 states, from New York City to San Diego, California. Learned to deal with the unexpected alone and to be ready to react and change plans as needed.
🎯 So Many Things Else
