Hire the Best Machine Learning Engineers
in China

Ni is a machine learning engineer at Cisco with an academic background in information engineering. SPECIALITIES: ✅Video Generation Models: VEO-3, seedance, wan, I2V, Stable Video Diffusion; ✅Generative AI Tools: VEO, GANs, Stable Diffusion (SD inpaint, SDXL), Flux, Fine-Tuning (DreamBooth, LoRA); ✅Programming Languages: C++, Python, JavaScript, Java; ✅Quality Control: LLM Evaluation & Benchmarking, Metrics-Driven Validation, AI-Powered QA Tools; ✅Deep Learning Tools & Libraries: PyTorch, TensorFlow, Hugging Face Transformers, ComfyUI; ✅Data Processing: NumPy, Pandas, OpenCV, SQL; ✅Model Training & Deployment: CUDA, Distributed Training, TensorBoard, Data Cleaning, AWS, Azure, Docker; ✅Certifications: AWS Certified Machine Learning – Specialty; ✅Awards: Silver Medal at the Kaggle Competition (Kaggle-LLM Science Exam); ✅Languages: Chinese (Native), English (Full Professional Proficiency, IELTS 6.5); Five years of experience in AI and software development represent a proven track record of delivering innovative AI-powered software applications. Through leading and participating in numerous projects, Ni is adept at leveraging data-driven insights to optimize algorithms and enhance product performance, also developed her expertise in cutting-edge technology solutions. With a mission to solve complex real-world problems and drive business success by pushing the boundaries of AI and machine learning, Ni continues further honing her skillsets to create innovative solutions, and is seeking new opportunities and challenges.

I build AI agents that survive production — LangGraph multi-agent systems, RAG that retrieves the right chunk, and MCP servers on live databases. Recent: a 6-agent booking platform on FastAPI + LangGraph handling flights, hotels and live payments. WHAT I BUILD - Multi-agent systems — LangGraph and LangChain orchestration, state machines, checkpointing, human-in-the-loop gates, multi-LLM routing across OpenAI, Claude and open models - RAG that actually retrieves — chunking strategy, embeddings, hybrid search, reranking, pgvector and Pinecone, plus an eval harness so you can prove retrieval improved instead of guessing - MCP servers — custom Model Context Protocol integrations connecting Claude and other clients to your databases, APIs and internal tools - Voice AI — streaming ASR and TTS over WebSocket, realtime conversational pipelines, built at the protocol level rather than wrapped around a hosted API - Agent + automation glue — n8n, Make and Zapier wired to real tool-calling agents, not just LLM prompt nodes - LLM optimization — prompt engineering, fine-tuning with LoRA/PEFT, and honest advice on when retrieval beats fine-tuning PROOF, NOT CLAIMS - smart_booking_agent — FastAPI + LangGraph platform coordinating six domain agents (flight, hotel, ride, ticket, planning, Q&A) with multi-LLM routing, Redis checkpointing, and live third-party booking and payment integrations including WeChat Pay v3 - xiaozhi-esp32-server — I lead this open-source voice AI assistant running on ESP32 hardware: WebSocket streaming, ASR, LLM dialog, RAG and function calling on a 240MHz chip. Publicly auditable. - Claude MCP server — production integration wiring Claude Code to a live database with scoped tool schemas and auth boundaries. 5-star reviewed on Upwork. - Fintech ML — credit scoring, recommendation and time-series forecasting models on noisy financial data. BACKGROUND M.S. in Pattern Recognition and Intelligent Systems, Beihang University (BUAA), China Scholarship Council scholar. Research: tree-structured Transformers for semantic parsing, outperforming baseline Transformers on three benchmarks. 5+ years shipping Python systems end to end — model to API to deployment. TECH STACK Python · LangGraph · LangChain · LlamaIndex · FastAPI · PyTorch · Hugging Face · OpenAI and Anthropic APIs · MCP · pgvector · Pinecone · Redis · PostgreSQL · MongoDB · Docker · n8n · Make HOW I WORK Based in Shenzhen (UTC+8) — I overlap comfortably with Asia, Australia, the Gulf and European mornings. I scope before I code, I tell you when something won't work, and I hand over documented, tested code you own outright. Send me the problem, not the spec. I'll tell you what it actually takes.

I am a Software Engineer at Bosch (Fortune 500). Previously at the Federal Reserve Bank of St. Louis, where I built RAG AI platforms used by 1000+ internal users, turning messy documents into structured data and insights. I was also a backend engineer at Siemens (Fortune 500), and MSAI at Carnegie Mellon University (number one in AI). At the Fed: • I indexed 1700+ documents and integrated 800k+ economic time series from the FRED and FRASER Database • Implemented hybrid RAG combines vector and keyword search (BM25&Trigrams) with Reciprocal Rank Fusion (RRF), • Designed data extraction pipelines to convert unstructured FOMC PDFs into structured defined schemas for querying, • Built multi-step workflows using LangChain/LangGraph, traceable and auditable actions with guardrails and citations, • Set up MCP server + API integrations, consistent access to structured data from internal systems with audit logs, • Delivered significant efficiency improvement, saving users up to 15 minutes per query. Currently at Bosch: • Developing Tessera, a conversation-driven data extraction system that converts unstructured documents into structured datasets via dynamic schema generation and iterative refinement At Siemens: • I developed Java backend services for a graph visualization platform, integrating with APIs and improving data processing throughput to support real-time visualization needs, matched performance of Neo4j on optimized workflows, • Built automated testing pipelines using GitHub Actions, enabling continuous integration for faster, more reliable releases, • Containerized services using Docker and deployed to AWS EKS, provisioning infrastructure with Terraform to ensure consistent, reproducible deployments At Institute of Automation, Chinese Academy of Sciences: • Deployed a computer vision traffic security system in Python and Pytorch supporting over 100k inferences per day through real-time data streams in a production environment. • Trained RepVGG, Yolo v5 models on 500k+ images for vehicle attribute detection, enhanced multi-task accuracy by 35% PUBLICATIONS Qufei Zhang, Yunshuang Wang, Gengsheng Li, Barry Cardiff, Pasika Ranaweera "Optimizing Federated Learning on Non-IID Data with Clustering and Model Sharing". EuCNC 2025 Jiahui Han, Qufei Zhang, Xiaoying Yang, and Jinyi Wang. "MIAE: A Mobile Application Recommendation Method Based on a Neural Tangent Kernel Model." IEEE BigData 2023 SKILLS Languages: Python, JavaScript, Java ML/AI: LangChain, LangGraph, RAG, RAGAS, Pinecone, OpenSearch, PyTorch DevOps: AWS, Docker, Kubernetes, Apprunner, Terraform, GitHub Actions, Supabase, Vercel Backend & Data: PostgreSQL, Kafka, MongoDB, Redis Frameworks: Flask, Node.js, Next.js, React, Django PS: The video on my page is for an AI lead follow-up app project I used to work on; you can get a sense of what I'm like, feel free to reach out if you wanna chat!

As part of a team, I bring extensive experience in problem-solving and critical thinking to the table. As a certified machine learning engineer and data scientist, I have worked on computer vision and deep learning applications and have been instrumental in creating architectures, training models, and turning data science prototypes into production-ready solutions. Additionally, I have a solid foundation in traditional machine learning approaches, conventional image processing techniques, and statistical analysis. Working with a team, I can contribute my skills in: • Deep Learning • Computer Vision • Image processing with OpenCV and Pillow • TensorFlow 2.x and PyTorch • CAD and Simulations • SolidWorks • Autodesk Fusion 360 • ANSYS • Data analytics • Data analysis • Data visualization using Tableau • Python • R • SQL • Excel • Powe BI • Spreadsheets • Machine learning algorithms • Statistical modeling • Hypothesis testing • Experimental design • Data preprocessing and cleaning • Feature selection and engineering • Model evaluation and selection Furthermore, I am adept at communicating technical concepts to both technical and non-technical audiences and believe in open communication, constructive feedback, and collaboration to achieve the team's goals.

🚀 Looking to integrate AI into your product but don't know where to start? Have you been told that AI can automate workflows, improve customer experience, and unlock new revenue streams—but you're unsure what is actually feasible for your business? Do you have valuable data but lack the expertise to turn it into actionable insights, predictions, or intelligent automation? Are you building a startup and need an engineer who can take your idea from concept to a production-ready web application without hiring an entire team? If any of these sound familiar, I can help. 💡 I specialize in building AI-powered software solutions that solve real business problems—not just research projects that never leave the lab. What I can help you build: ✅ AI Chatbots & LLM Applications - Customer support assistants - AI-powered search and retrieval systems - Workflow automation agents ✅ Machine Learning Solutions - Custom computer vision models - Recommendation systems - Predictive analytics ✅ Full-Stack Websites - React, Next.js, Typescript in frontend - Python, Django, FastAPI backend - MySQL, Postgres, MongoDB for database - End-to-end product development Whether you need to add AI to an existing product, outsource part of your machine learning pipeline, build an MVP for investors, or develop a complete AI-powered platform, I can help turn your idea into a scalable solution. 📞Get a feel for yourself by sending me an invite or scheduling a call with me.📞