Hire the Best Data Scientists
in China

Hello~ I'm Yuanhao with Master degree plus 4 years in AI product development. Now working as Data Scientist in Unilever. Mainly responsible for the development of NLP and CV related tasks. Proficient in Python, Azure and mastering basic SQL, Linux, C++. Interesting in embedded development such robots and drones. Daily interests are sports and photography

I build and evaluate production AI agents for real business workflows. I work from ambiguous requirements through architecture, implementation, evaluation, deployment, observability, and handoff. Recent delivery: I completed Kinetix Mentor through a $5,600 / 140-hour Upwork engagement. It is a deployed WhatsApp AI platform combining Gemini models, LangGraph orchestration, Pinecone RAG, PostgreSQL memory, multimodal processing with GCS, Stripe subscriptions, external research tools, LangSmith tracing, FastAPI, and Docker. The client gave the project 5 stars and described me as “a complete professional who gets the job done.” I am now turning Kinetix into a measurable Agent Evaluation system. Current work includes: • Versioned JSONL evaluation datasets and repeatable model runners • Deterministic scoring for tool selection, routing, arguments, security, and response contracts • Repeated live-model evaluation instead of one-off testing • Trace-driven failure analysis and targeted regression testing • Separation of scorer defects, agent defects, and infrastructure failures The first router baseline evaluated 30 repeated model runs, uncovered both measurement defects and genuine agent failures, and produced a 17/17 passing targeted regression after fixes. I can help you: • Build or repair LangGraph, RAG, and tool-using AI agents • Create evaluation datasets, behavioral scorers, and regression suites • Diagnose production traces and improve reliability, quality, latency, and cost • Integrate FastAPI, PostgreSQL, vector databases, webhooks, payments, and external APIs • Turn an AI prototype into a deployed, observable workflow with clear documentation and handoff Before freelancing, I worked in ByteDance’s international business, building data-driven systems and translating ambiguous business problems into measurable technical outcomes. Best fit: AI Agent and RAG systems, LLM evaluation, existing AI product improvements, reliability audits, and production integrations. Send me your workflow, current architecture, failure examples, and success criteria. I’ll propose a concrete first milestone.

1. more than 8 years experience in data analytics field 2. worked 10 years in international projects, fluent communication in English & German 3. good knowledge in the following technologies, including Tableau, PowerBI, Python, R, SQL, Excel

I am a Ph.D. student and AI researcher with a strong focus on deep learning, computational neuroscience, and Large Language Models (LLMs). Working daily in a high-paced research lab, I specialize in transforming raw data into reproducible, high-performing machine learning pipelines. Whether it's fine-tuning LLMs for specific downstream tasks or building complex computer vision models, I bring rigorous academic standards and engineering best practices to every project. Core Tech Stack & Skills: Deep Learning Frameworks: PyTorch (Primary), TensorFlow, Keras. LLMs & NLP: Model fine-tuning (LoRA, PEFT, Hugging Face transformers), sequence classification, and prompt optimization. Computer Vision & Medical AI: 3D/2D CNNs, Vision Transformers (ViTs), U-Net architectures, and neuroimaging (MRI) analysis using MONAI, SimpleITK, and nibabel. Data Science & MLOps: Python, Pandas, NumPy, Scikit-learn. Experiment tracking (W&B, TensorBoard), version control (Git), and environment management (Docker, Conda, Linux)

I’m a data scientist with an MSc in Scientific Computing from Durham University, specializing in structured data modeling, time series forecasting, and interpretable machine learning. My work focuses on extracting meaningful structure from complex datasets and turning it into reliable, usable models — not just black-box predictions. I have experience in: - Time series forecasting (retail demand, sales prediction) - Feature engineering for tabular data - Machine learning with LightGBM and Random Forest - Interpretable modeling (e.g. symbolic regression) Selected projects: • Retail Sales Forecasting Built a time series model using feature engineering and LightGBM. Identified key drivers such as weekly seasonality, promotion effects, and transaction-based demand signals. Achieved robust performance with proper time-based validation and leakage control. • Interpretable Stellar Mass Estimation Developed an interpretable ML model that recovers a simple analytical formula from observational data, achieving high accuracy while maintaining full transparency. I’m comfortable working from messy real-world data to a structured solution, including: EDA → feature engineering → modeling → evaluation → clear reporting If you're looking for someone who can not only build models but also explain and structure the problem clearly, feel free to reach out.

I build ML and AI systems that ship — and I design the evaluation and experiments to prove they work in production. My work spans recommendation systems, forecasting, churn and ETA prediction, and more recently LLM applications including RAG and AI agents. 11 years of production ML at Meituan, Didi, and LINEMAN (Thailand's largest food-delivery platform). Recent results: • Delivery-time (ETA) model — XGBoost with quantile loss, cut long-tail MAE by 6.8% (A/B tested) • Multi-objective ranking (MMoE) + DPP re-ranking for a B2B marketplace — +3.87% GMV per user, +6.53% CTR • RAG question-answering system over 10,000+ company financial reports — embeddings + FAISS + LLM What I can build for you: • LLM apps & RAG: document Q&A and chatbots grounded in your own data (OpenAI / Anthropic / open-source models, FastAPI backends, vector search), with an eval set so you know it works • AI agents & workflow automation: tool-calling agents, structured output, guardrails • Forecasting & predictive modeling: demand, ETA, churn — explainable and A/B-tested • Recommendation systems: retrieval + ranking + re-ranking, cold start, latency budgets Stack: Python, SQL, PyTorch, Spark, FastAPI, LangChain, AWS SageMaker. Based in UTC+8 — I overlap US mornings and EU afternoons.