Ying Zhou

MS in Computer Science

2010 - 2014

Relevant Coursework

• Deep Learning
• Computer Vision
• Foundation Of Algorithms
• Wavelet
• Python

BS in Computer Science

2014 - 2016

Relevant Coursework

• Mathematics
• Algorithms & Optimization for Big Data
• Machine Learning
• Computer Science
• Matlab/Mathematica

INNOS

January 2024 - Present

Senior Software Engineer

• Built and scaled a SaaS platform with microservices architecture to support multiple LLM agents for personalized content generation, writing assistance, and intelligent search, using ChatGPT, Claude, Llama, DeepSeek, and LangGraph/CrewAI frameworks.
• Integrated multimodal capabilities including CLIP, BLIP, Whisper, and ElevenLabs, enabling LLMs to analyze PDFs, images, tables, and voice input, with support for RAG, KAG, and CAG pipelines using vector databases for enhanced contextual responses.
• Developed secure, decentralized features using FastAPI, Node.js, Solidity, and IPFS/Filecoin, including on-chain metadata linking and autonomous content generation for social platforms, with responsive Web3 frontends in React/Next.js.
• Implemented and customized Microsoft 365 Copilot & Copilot Studio to automate business processes across Word, Excel, Outlook, and Teams, building low-code workflows with Power Automate, Dataverse, and SharePoint integrations.
• Designed and maintained CI/CD pipelines with Docker, Kubernetes, Terraform, and led agile sprints, human evaluation feedback loops, and performance tracking dashboards (Tableau) to continuously optimize LLM agent accuracy, latency, and engagement.

NVIDIA

July 2022 - Jan 2024

AI Strategy Engineer (Technical Leader)

• Built a real-time patient monitoring system using Transformer-based models, BlazePose, LSTMs, and Temporal Fusion Transformer, achieving 98% accuracy in multi-label activity classification and detecting stroke, seizures, and Parkinson’s tremors from video and IoT sensor data.
• Optimized models (ViT, MobileViT) for edge inference at 25 FPS using TensorRT, ONNX, and knowledge distillation, and integrated emergency response via Twilio and FCM, cutting alert time to under 30 seconds.
• Developed HIPAA-compliant backends with FastAPI/Django, deployed on Google Cloud (Vertex AI, BigQuery) and Azure, and containerized with Docker/Kubernetes for low-latency inference and scalable integration into hospital systems.
• Built hybrid AI pipelines combining deep learning (PyTorch, TensorFlow) and traditional ML (XGBoost, SVM, scikit-learn), automated with Apache Airflow/Kubeflow, and processed large-scale streaming data via Apache Spark and Databricks.

July 2018 - July 2022

Industrial Machine Learning Engineer

• Built a ResNet50-based object detection model for real-time defect detection in manufacturing, achieving 92% precision, 94% accuracy at 30 FPS, enhanced by data augmentation and Wiener filtering for noise reduction.
• Developed autoencoder-based image compression and real-time data pipelines using Apache Kafka, enabling efficient processing of industrial images and sensor streams.
• Led end-to-end MLOps workflows with Docker, Kubernetes, and automated retraining (via cron jobs), deploying models with blue-green and canary strategies, and ensuring seamless collaboration with backend and DevOps teams.

SenseTime

July 2019 - Dec 2019

Software Developer

• Review the state-of-the-art traditional image denoising algorithms and results to lay the foundation for development of a novel approach.
• Designed the main algorithm structure using sparse representation(long-tailored distribution), self-similarity regularization and multi-scale processing techniques to preserve critical image details such as textures and edges.
• Developed Image Denoising Project by using this algorithm, Flask, Restful APIs, asynchronous functions enabling efficient handling of image processing tasks.
• Experimented with various configurations and tuning parameters to maximize the PSNR performance.
• Achieve superior PSNR scores compared with state-of-the-art denoising algorithms available at that time, demonstrating better image quality and noise suppression.
• Contributed to research, documenting using the IMRAD(Introduction, Methods, Results, and Discussion) format for the algorithm’s enhancements and benchmarking results, which were presented to the publications.

Languages and Databases

Frameworks

Tools