Houhan Lu Email & Phone Number

New York, Ny, Us

New York, United States

- Led the development of a full-stack web application for a label company, including backend development with Flask and Amazon DynamoDB, and frontend development with JavaScript, jQuery, and Bootstrap.- Managed production and development servers on AWS EC2, setting up PDF and document storage on AWS S3.- Developed real-time chat, voice call, and notification system using Socket.IO server for seamless communication.- Refactored and optimized PostgreSQL database operations with SQLAlchemy, enhancing performance and scalability.- Built end-to-end ML pipelines: containerized models with Docker, exposed models as a service using Flask, registered Docker images with Harbor, and deployed containers to Kubernetes using Rancher.- Enabled customizable ML pipelines for data ingestion, preprocessing, model training, and inference on AWS.- Utilized GPT-3.5 Turbo for prompt engineering to enable SQL query execution for users without SQL knowledge.- Conducted sentiment analysis using VADER and keyword extraction with TextRank on comments, aggregated to artist level.- Researched and prototyped generative AI use cases for Named Entity Extraction (NER) using LLMs like Hugging Face Transformers, LangChain, Llama, GPT-3.5 turbo, and BERT.- Provided business insights from demographic, behavioral, and sentiment data on social media to enhance marketing and advertising performance.

New York, United States

- Developed deep learning models using CNNs, RNNs, Transformer encoders, and large-scale pre-training.- Built Masked-Auto Encoder models for X-ray and health records representation using the MIMIC dataset.- Created a CLIP-like vector embedding model, integrating text and image representations, optimizing model performance to achieve a 95% precision score.- Developed a novel Seq-Seq model combining Marginal Structural Models and Representation Learning for time series analysis, achieving a 0.6% RMSE and a 48.1% performance improvement.- Built machine learning models (Logistic Regression, Random Forest, Gradient Boosting) to predict hospital visit probabilities.- Developed predictive models to suggest medicines for patients with complex medical histories.- Created an expert system in Python using RDKit and DoWhy to predict drug and excipient compatibility with 97% accuracy.- Researched and implemented statistical methods to extract meaningful descriptors from fMRI datasets of patients.

Beijing, China

- Collaborated with cross-functional teams remotely, using Git for version control.- Developed a novel Seq-Seq model combining Marginal Structural Models and Representation Learning for time series analysis, achieving a 0.6% RMSE and a 48.1% performance improvement.- Proposed a masked diffusion method to address gradient leakage, reducing SSIM losses by 18.2%.- Finetuned the GPT-3.5-turbo to extract root causes of customer calls in the healthcare domain, using labeled data from GPT-4 via prompt engineering and few-shot learning.- Implemented a denoising autoencoder, achieving a 22.31% reduction in noise for noisy genomics data.- Conducted ablation studies on CNN, RNN, GRU, and stacked autoencoders to fine-tune models for binary chromatin data.

Chengdu, Sichuan, China

- Developed ETL data pipelines using a cloud-native tech stack, leveraging Python, SQL, Docker, and AWS services (S3, Lambda).- Automated data processing workflows, creating scalable software for pipeline triggering, monitoring, and validation using AWS Batch, CloudWatch, and Splunk.- Developed CV models to detect and classify medical components in power station environments, reducing toxic chemical risk.- Deployed an inference engine to run 3D sparse convolutions and custom CUDA kernels in under 30ms, achieving an 85% speedup from PyTorch, facilitating fast model inference on location of lightning strokes.

Kunming, Yunnan, China

- Created and analyzed Hive tables; implemented partitioning, dynamic partitions, and buckets in Hive.- Enhanced data processing pipeline and CTR algorithms (LR, GBDT+LR, DNN) using LightGBM and XGBoost for massive datasets on a big data platform.- Automated data processes using tools like Tableau, D3.js, Shiny, and Spark.- Developed ML and computer vision algorithms for retail inventory management systems.- Designed a ML algorithm with ~93% accuracy to target specific user groups for surveys and revenue generation.- Increased model prediction querying rate by 500x through tuning and parallelizing Spark transformers.- Integrated backend functionality and APIs using Node.js, reducing API response time by 12.7% by minimizing database queries.- Optimized Django ORM queries with pre-fetching and RawSQL, reducing database hits by 23%.- Managed administrator and user identity through Auth0 Authentication API.- Developed a labeling tool using GRPC in Golang and Python, streamlining processes for a team of 50 labelers.- Designed scalable data ingestion architecture and feature pipelines using Spark, Hive, PostgreSQL, and SQL Server.- Built a BERT-based representation learning model for query, product, ad, and brand to enhance cold-start and coverage performance.- Developed an NLP-based system to generate product descriptions for a retailer using product attributes.- Trained and fine-tuned a transformer-based model, using regex, POS tagging, and dependency parsing for data preprocessing.- Conduct exploratory data analysis (EDA) to gain insights into network patterns and trends.

Chengdu, Sichuan, China

- Designed and implemented a multi-modal deep learning network using Camera and Lidar data for ground height and lane estimation.- Developed a 3D PointNet model for temporal smoothing of segmentation predictions over point cloud sequences.- Built a Sparse Point-Voxel CNN for semantic segmentation of point cloud sequences, increasing data annotation speed by 30%.- Evaluated the accuracy-latency trade-off of object detection models including YOLOv4/v5 and EfficientNet.- Developed a website for an experiment with over 100 subjects to compare topological visualization techniques using Node.js and Three.js; generated mesh model data for Reeb Graphs and persistence diagrams with Python.- Built a full-stack web application for an Online 3D Resource Database using Python, React, JavaScript, JQuery, HTML, and CSS.- Implemented a dashboard to display 3D data from REST endpoints in JSON format, facilitating easy navigation, preview, and manipulation of object materials and shaders.- Collaborated with UI/UX designers, Product Managers, and Business Analysts to ensure optimal user experience and functionality.

Chengdu, Sichuan, China

- Developed OCR solutions for document digitization using machine learning and deep learning, training models on the EMNIST dataset for handwritten text/number recognition.- Conducted marketing analytics projects using R and SAS, including pre-launch planning, target prioritization, control group identification, and post-launch ROI analysis.- Worked on back-end query processing, data mining, ETL, data integration/migration, and data flow creation/modification.- Deployed Flask applications to enable physicians to make data-driven decisions quickly, enhancing patient treatment.- Conducted data analysis on Vancomycin, achieving a 30% reduction in infection rates in a clinical study.- Developed a predictive model to optimize Vancomycin dosage for infections, resulting in a 25% improvement in recovery time.- Led a team to perform sensitivity analysis on Vancomycin dosage and treatment duration, identifying optimal ranges and improving infection reduction effectiveness by 40%.

Master Of Science - Ms, Computer Science

Education record