LLM Engineer / Biomedical Applied Scientist
Shenzhen / Shandong / Hong Kong / Remote
Goal: Learn from data, grow through training, live with aligned values—cycle evolves.
Real: Messy data, failed training, misaligned goals, oops. 😓
Open to full-time roles
A unified, intelligent portal that integrates complex tools, enabling researchers to complete the entire process from image analysis to performance prediction through natural language conversation. This is my graduation project—AI-Fiber—an intelligent research assistant for electron microscopy and electrospinning based on large language models (LLMs).
This project demonstrates the deep learning framework of SpinSci-Ex, designed to automatically extract structured text such as manufacturing parameters, morphological features, and functional properties from scientific literature.
The core of this project is an end-to-end intelligent system consisting of two parts: a YOLO-OCR model for automatically extracting scale bar information, and an interactive measurement frontend powered by MicroSAM that integrates this information.
Designed and implemented the SAM-Fiber (Segment Anything Model for Fiber) framework. It goes beyond simply applying generic models by introducing a series of innovations to adapt the powerful visual foundation model (SAM2) into a specialized fiber analysis tool.
This project utilizes wireless devices to create a fabric image dataset containing detailed structural parameters, and proposes a multi-task multi-scale convolutional neural network (MTMSnet) to identify fabric density and weave patterns, achieving errors below 2% and 10% respectively.
The client's existing SaaS service was inefficient. To enhance user experience, there was a need to provide an intelligent AI assistant available 24/7.
· Responsible for using Dify for knowledge base construction, application development, and prompt orchestration. · Led the organization and upload of over 100 product-related technical documents, establishing an authoritative knowledge base. · Evaluated multiple large models, selecting the optimal model based on subjective assessment, response speed, and API cost.
Deployed the first version of the AI Q&A system within just one week, reducing repetitive inquiries by approximately ⬇️40% and achieving a user satisfaction rate (based on feedback buttons for answer usefulness) of ⬆️75%.
Mar 2025 – May 2025 · Dify, Prompt Orchestration, Workflow, Knowledge Base
To address the issue of fragmented and difficult-to-retrieve historical achievement information, a modern knowledge base platform for innovation outcomes needed to be built.
· Designed and implemented a content management system, transforming thousands of unstructured invention patents and project records into structured knowledge entries with metadata tags. · Independently handled front-end and back-end development, cloud server deployment, and designed scalable APIs for future AI integration, laying the data and interface foundation for subsequent intelligent Q&A (RAG) implementation.
Successfully digitized over 3,000 invention achievements, improving information retrieval efficiency by ⬆️90% after the platform launch.
Jul 2024 – Sep 2024 · Structured Data, Knowledge Base Construction, API Design, Full-Stack Development, RAG Foundation
As a startup team, there was a need to build a cross-platform financial system from scratch to achieve financial data visualization and management.
· Led front-end architecture selection, using Uni-app, Vue, and ECharts to build a visual interface enabling dynamic interaction with multi-dimensional financial reports. · Collaborated closely with back-end engineers to successfully deploy one codebase to both web and WeChat mini-program platforms.
After the system launch, report verification efficiency improved by ⬆️50%, achieving visualization of key data.
Jan 2019 – Jun 2023 · Vue, Uni-app, ECharts
To integrate the company's fragmented production, sales, and inventory data, a new ERP system was needed to improve management efficiency.
· Responsible for back-end development of production scheduling and inventory management modules. · Built cross-service communication interfaces using the SSM framework and Dubbo technology, enabling efficient data synchronization between order and warehouse systems.
Optimized SQL queries and indexes, improving the generation speed of key reports by ⬆️30%.
Jan 2018 – Jun 2018 · Java, Spring MVC, Mybatis, Database Design
2025-08-29 · 1 min read
The project adopted a wireless device and established a fabric images dataset with elaborated structure parameters. Presented a multi-task and multi-scale convolutional neural network (MTMSnet) to recognize the fabric density, weaving pattern, reaching under 2% and 10% error respectively.
2025-08-29 · 3 min read
Is it possible to integrate these complex tools into a single, intelligent entry point, allowing researchers to complete the entire process from image analysis to performance prediction using only the most natural form of interaction—language conversation?
2025-08-28 · 2 min read
This project designed and implemented the SAM-Fiber (Segment Anything Model for Fiber) framework. Instead of simply applying a general-purpose model, it transforms the powerful vision foundation model (SAM2) into a specialized fiber analysis tool through a series of innovations.
2025-08-28 · 3 min read
The core of this project is an end-to-end intelligent system composed of two parts: a YOLO-OCR model for automatically extracting scale information, and an interactive measurement front-end powered by MicroSAM that integrates this information.