Data Scientist · Lyft
Leading optimization and machine learning systems for real-time driver incentives, earnings prediction, and guarantee forecasting.
Rachel Yu
I build optimization, machine learning, and real-time AI systems.
Currently a Data Scientist at Lyft.
Previously a Senior Associate Data Scientist at NTT DATA.
Experience
(5)Building intelligent systems from research through production.
Senior Associate Data Scientist · NTT DATA
Built edge AI, multi-agent RAG, computer vision, and full-stack 3D object detection systems for production use cases.
Graduate Research Assistant · University of Calgary
Researched real-time UAV obstacle avoidance, small-object detection, and path planning on an 11TB+ imagery dataset.
Software Developer · GNO-SYS Technology
Deployed Pangeo on AWS and Kubernetes, orchestrating and scaling 128 Docker containers.
Software & Test Engineer · BaiDu
Improved a Jetty-based signup service and its test coverage, cutting request time by up to 30%.
Projects
(5)High-impact systems across optimization, generative AI, and computer vision.
Daily Stock Analysis Agent
A multi-agent stock research pipeline combining technical, news, quantitative, and risk agents with a human approval checkpoint.
Real-time driver incentive optimization
Led a cross-functional 0-to-1 launch of low-latency incentive optimization services driving $30M+ in annual profit.
Real-time 3D object detection
Designed custom inference algorithms and modular streaming APIs for a patent-filed real-time detection system.
Recorded project demonstration
PPE safety-zone detection
Detected people and safety helmets with YOLO, estimated each person's position from a monocular camera using geometric algorithms, and applied polygonal safety zones to highlight workers inside restricted areas without helmets using red bounding boxes.
UAV obstacle avoidance with Circle RRT*
Combined object detection with classic pathfinding to reduce computation time by 20% and path length by up to 6%.
Publications & Patents
(5)Research in UAV perception, path planning, responsible innovation, and computer vision systems.
Monocular Vision Measurement for UAV Obstacle Avoidance and Path Planning with Circle RRT*
Shaping Tomorrow: Responsible Innovation for a Brighter Future
Two-Stage Approach to Small-Object Detection
Small-object Detection for UAV-based Images
UAV path-planning simulation
System and Method for Class-Agnostic Counting of One or More Items in a Container
Learning docs
(2)Interactive notes I build while learning ML systems.
Where each LLM speed and memory technique fits
An interactive pipeline map of 40+ LLM speed and memory techniques, plus a VRAM and decode-speed lab for an 8B model.
Agent Harness
An interactive walkthrough of a LangGraph run and the architecture around a production agent harness.
Technical toolkit
Tools I use to build, validate, and ship.
AI / ML
LLM fine-tuning (LoRA), quantization, LangChain, LangGraph, RAG, PyTorch, TensorFlow, CatBoost, computer vision
Languages
Python, TypeScript, JavaScript, Java, C/C++, SQL, NoSQL, Shell
Backend & Infra
FastAPI, Flask, Django, Express.js, Docker, Kubernetes, AWS, Azure, Terraform, CI/CD, Spark
Frontend
React, React Native, Next.js, Tailwind CSS
Education
A foundation in software engineering and intelligent systems.
Master of Science in Software Engineering
GPA 4.0/4.0 with a full scholarship.
Bachelor of Engineering in Electronic Information Engineering
Learning docs
(2)Interactive notes I build while learning ML systems.
Where each LLM speed and memory technique fits
An interactive pipeline map of 40+ LLM speed and memory techniques, plus a VRAM and decode-speed lab for an 8B model.
Agent Harness
An interactive walkthrough of a LangGraph run and the architecture around a production agent harness.