Yiming Fu

AI Engineer at Mahanova (MiraclePlus, formerly YC China-backed startup)

RAG Development: Built and optimized an end-to-end LLM-powered knowledge assistant, covering intent recognition, query rewriting, web retrieval, and response post-processing (attribution & compliance filtering). Document Parsing: Led development of a hybrid document parsing module (OCR + structured metadata extraction) for multi-modal inputs (PDF, Excel, etc.), enabling scalable knowledge base construction. Agent Prototyping: Contributed to early-stage LLM agent assistant development (LangChain, memory management), supporting tool use and multi-step reasoning workflows. Testing & Deployment: Responsible for code debugging and unit testing using Jira/Confluence, Git, Docker, and Jenkins for CI/CD.

Software Engineer Intern at Shutterstock

API Development: Backend development of the media asset copyright review status function, including SQL database creation and modification, backend logic implementation using Java and Spring Boot frameworks, and REST-style API development and management based on Swagger.

Researcher at The University of Manchester

Cognitively Inspired Benchmark for Interpreting Numerosity Representations in Vision–Language Models; ECCV 2026 (1st author).

Researcher at The Chinese Academy of Sciences

PGT-Net: Prior Guided Transformer for Hyperspectral Image Denoising; IEEE Multimedia 2025 (2nd author).

Researcher at The Chinese Academy of Sciences

Model Fine-tuning: End-to-end pipeline for processing 240,000 scanned pages; fine-tuned vision-language models (e.g., OVIS) using SFT and LoRA for fine-grained image classification, achieving ~91% accuracy.

Researcher at Hong Kong University of Science and Technology

Using Machine Learning Method to Assist Rare-book Edition Identification; Digital Humanities 2025 (2nd author).

AI Engineer at Mahano va (MiraclePlus, formerly YC China-backed startup)

Built and optimized an end-to-end LLM-powered knowledge assistant (RAG) including intent recognition, query rewriting, web retrieval, and response post-processing (attribution & compliance filtering). Led development of a hybrid document parsing module (OCR + structured metadata extraction) for multi-modal inputs (PDF, Excel, etc.), enabling scalable knowledge-base construction. Contributed to early-stage LLM agent development (LangChain, MCP, memory management), supporting tool usage and multi-step reasoning workflows.

Researcher at Tsinghua University

Supporting Children's Personal Emotion Education through Parent-AI Collaboration; 1st author, work under review.