Machine Learning Ann Arbor

Yiyang Fan

805-***-**** **********@*****.*** https://www.linkedin.com/in/yiyang-fan-80a3ab299/

EDUCATION

University of Michigan Ann Arbor, MI 08/2023-05/2026 Bachelor of Science in Computer Science; Minor in Mathematics, Business. GPA: 3.88/4.00 Relevant Courses: Web Systems(A-), Data Structure & Algorithms(A-), Machine Learning(A), Marketing Management(A-) SKILLS

Programming: Python, C++, Java, Unix/Linux, JavaScript, HTML/CSS, Flask, Jinja2, MATLAB, LaTeX ML & AI: LLMs, RAG, TextGrad, generative AI, NLP, Reinforcement Learning (PPO), Monte Carlo Tree Search, Regression, Random Forests, Boosting, Neural Nets (CNNs, RNNs, Transformers), GMMs, Clustering, Transfer Learning Frameworks & Tools: PyTorch, Jupyter, GitHub, Docker, Ollama, Chromadb, React, Flask, MapReduce, GFS Infrastructure: AWS (EC2, Load Balancing), Azure, Docker, Distributed Systems, Sharding/Replication, DNS/CDN Databases & Data Analysis: SQL, MongoDB, JSON, Excel, R, Data Visualization, PageRank, Statistical Modeling Systems: ARM, pipelining, hazard handling, memory-mapped I/O, virtual memory, caching (LRU, set-associative) Concurrency: Multithreading, Async I/O; Engineering Tools: Fusion 360, Onshape, SolidWorks, AutoCAD INTERNSHIP EXPERIENCE

Developer Intern - Information and Technology Services, University of Michigan Ann Arbor, MI 05/2025-08/2025

● Developed and deployed a full-stack study spot recommendation system with LangGraph frontend and AWS Lambda backend, enabling real-time location and environment-based queries

● Designed and benchmarked multiple retrieval pipelines—Basic LLM Query, DSPy, and Contrastive Reranker + Hybrid RAG—achieving accuracy/time trade-offs (e.g., 68.06% @ 52.16s, 52.06% @ 12.75s, 59.72% @ 28.02s)

● Implemented caching mechanisms to reduce query latency, improving system responsiveness

● Conducted user surveys for initial requirements gathering and post-deployment validation; iterated on architecture based on stakeholder feedback by modularizing location and study spot components for better reusability AI Engineer Intern - C12.ai Shanghai, China 06/2024-08/2024

● Integrated RAG pipelines using various LLMs, embedding, and retrieval models—including TextGrad and Monte Carlo Tree Search—to automate chemical process analysis, increasing accuracy from 60% to 75%

● Refined and validated outputs with chemical technicians; iteratively improved 50+ JSON-based test sets

● Proposed parallel workflow to continue testing while drafting model proposals after boss paused test phase PROJECT EXPERIENCE

Paco’s Puppy Problem – Dog Breed Image Classification PyTorch, CNN, ViT, Transfer Learning 02/2025-03/2025

● Built a robust classifier distinguishing Golden Retrievers from Collies using 12K+ labeled images across 10 breeds.

● Implemented CNNs and ViTs from scratch; achieved 0.8512 AUROC using transfer learning and a custom CNN.

● Pretrained on 8 auxiliary breeds; layer freezing improved generalization (AUROC from 0.75 to 0.8136).

● Tuned hyperparameters (SGD/Adam, batch size, weight decay) and used label smoothing, random erasing.

● Compared CNN vs. ViT; ViT underperformed (AUROC = 0.5524) due to low inductive bias on small data. Wikipedia Search Engine Python, Hadoop, Flask, React, MapReduce 04/2025-04/2025

● Built a multi-stage Hadoop pipeline using Python streaming to generate tf-idf–weighted inverted indexes.

● Implemented document normalization and a custom splitter to enable parallel processing of 3,000+ articles.

● Integrated PageRank into scoring to improve result relevance.

● Deployed a Flask REST API (/q=<query>&w=<weight>) returning top-ranked results as JSON.

● Built a React frontend with search bar, result panel, and a tf-idf/PageRank weighting slider (0–1). Instagram Clone Projects (Insta485 Series) Python, Flask, React, SQLite, AWS 01/2025-02/2025

● Built a Python static site generator using Jinja2 and JSON, rendering Instagram-like pages with templated HTML/CSS.

● Developed a Flask app with SQLite backend and 15+ routes supporting full CRUD, session auth, and secure file uploads.

● Designed SQL schema with ON DELETE CASCADE; added hashed passwords, UUID media, and access control.

● Deployed Flask server on AWS EC2 with Nginx, verified endpoints with curl and end-to-end integration tests.

● Refactored to REST API and React frontend with Hooks, Fetch, infinite scroll, and real-time like/comment updates.

● Added async UX (double-tap to like, dayjs timestamps), bundled via webpack, tested with Cypress and deployed to AWS. MapReduce Framework Python, TCP/UDP Sockets, Multithreading, Distributed Systems 03/2025-04/2025

● Designed a Manager-Worker architecture using TCP/UDP sockets for job distribution and fault-tolerant coordination.

● Implemented TCP protocol for task assignment and result transfer; used UDP for Worker heartbeats and liveness checks.

● Built multithreaded server to handle concurrent TCP clients and listen for UDP heartbeats asynchronously.

● Partitioned input, assigned map tasks round-robin, hashed keys for reducer grouping, and merged final outputs.

● Enabled fault tolerance by reassigning tasks after 5 missed heartbeats; validated reallocation via Worker termination tests. LEADERSHIP EXPERIENCE

Outreach Chair - Girls Who Code Loop @ UM Ann Arbor, MI 11/2023-Present

● Led a career exploration event with Girls Group: developed content, recruited guest speakers, arranged venue, managed budget, and handled event logistics. And promoted the event through student channels, increasing attendance by 40%

● Conducted Python workshops for 50+ high school students in collaboration with Laguna Creek High School, enhancing hackathon skills and boosting program engagement by 27%

Contact this candidate