About Me

AI Application Development Expert with 9 years of experience, specializing in RAG and Agent technologies. Dedicated to building cutting-edge AI solutions and driving knowledge sharing through technical writing.

Work Experience

Acao (HK) Limited

AI Engineer

2025.6-2025.8

Custom RAG Solution for Enterprise Document Intelligence

Architected and delivered a custom RAG system to automate Q&A from complex PDFs for a client, slashing manual data lookup time by 90%.
Spearheaded client requirement workshops, designing a solution with Paddle OCR parsing pipeline, an LLM-powered knowledge base, and a hybrid ColBERT/Elasticsearch retriever to maximize answer accuracy.
Engineered a cost-optimized deployment with open-source model fallbacks, ensuring 99%+ uptime while reducing operational costs.

Multilingual Speech-to-3D Model System for Education

Engineered an end-to-end system converting Cantonese, Mandarin, and English speech into 3D-printable models, driving innovation for educational products.
Architected a modular pipeline integrating speech recognition (Baidu API), semantic understanding (Qwen LLM), and 3D asset generation (Meshy API) for seamless workflow automation.
Enhanced learning interactivity by providing a tangible, voice-driven creation tool, significantly boosting student engagement and practical application of knowledge.

Shenzhen Chinasoft International Tech Services Co., Ltd.

Senior NLP Developer

2018.4-2024.4

Core Developer - Huawei Financial AI Platform

Spearheaded the intelligent financial Q&A system; led the model upgrade from Word2Vec to BERT and implemented a hybrid ranking algorithm, increasing answer accuracy to over 95% and reducing the repeated question rate by 30%, serving 8,000+ daily active users.
Built a financial knowledge graph by applying an NER model to resolve entity boundary ambiguity; constructed a graph with 150k+ entities and 500k+ relationships on Neo4j, boosting query efficiency by 60% via an incremental update mechanism.
Developed a Text-to-SQL system optimized for complex queries, achieving 92% SQL generation accuracy and significantly improving the finance department's data query efficiency.
Implemented a multimodal document Q&A system integrating PaddleOCR for layout analysis and a vector database for retrieval, achieving over 90% answer accuracy and processing tens of thousands of documents monthly.

Algorithm R&D - Huawei General AI Platform

Fine-tuned LLaMA and ChatGLM models for the financial domain using LoRA technology, enhancing their performance on specialized tasks and training efficiency.
Established a large model content safety system capable of toxic content detection (96% accuracy) and hallucination detection (85% accuracy), processing tens of thousands of entries daily.
Designed an image data drift detection solution combining KL-divergence statistical analysis with a VAE model; introduced active learning, which increased annotation efficiency by 40% and achieved a final detection accuracy of 94%.

Beijing Huibao Haitai Network Tech Co., Ltd.

NLP Developer

2016.12-2018.3

Insurance Domain Intelligent Q&A System (Simba) & Data Platform Construction

Spearheaded the development of an intent recognition module utilizing Word2Vec, cosine similarity matching, and keyword rules, and constructed a dedicated insurance knowledge base. Achieved over 85% accuracy in internal tests, significantly enhancing customer service efficiency.
Led the migration from MySQL to PostgreSQL, optimizing indexes and table structures to significantly reduce query latency and improve system stability and concurrent performance.
Built an end-to-end automated ETL process using Python and Kettle, achieving full-link automated data processing, substantially reducing manual effort, and ensuring data consistency and processing efficiency.