AI ENGINEERING PROGRAM TRAINEE
VNET Group
Working on enterprise AI applications, RAG systems, AI agent workflows, data analytics, and front-end product design for data-center operations.
- Built an AI operations and project collaboration prototype in Frappe, implementing three application modules, 11 DocTypes, and 27 API methods for internal AI requests, meeting records, task participation, project progress, and audit trails.
- Integrated an OpenAI-compatible LLM client with rule-based fallback and human review. Implemented rule-based duplicate screening across three text fields and row-level access controls for five document types.
- Deployed a four-container RAG prototype with Docker Compose and verified indexing of 15 sample documents across seven knowledge bases. Evaluated retrieval with 14 queries over three PDFs: hybrid retrieval returned a relevant document in the top three for 12 of 14 queries (85.7%), compared with 11 of 14 (78.6%) for keyword search.
AI Operations & Project Collaboration
A Frappe prototype for internal requests, meeting records, task participation, project progress, and audit trails.
Rule-based duplicate screening covers three text fields. Row-level access controls cover five document types. The prototype was built for a 2,000+ employee enterprise.
Enterprise Multimodal RAG Platform
Collaborated on front-end design, deployment, and validation of a private enterprise multimodal RAG platform.
Designed user workflows across knowledge bases and sample documents. Tested parsing, chunking, indexing, and hybrid retrieval; investigated Chinese retrieval, embedding, and LLM-service failures across the pipeline.
Hybrid vs. keyword
14 queries · 3 PDFs · 407 chunks
Top-3 document retrieval hit rate
Small retrieval evaluation; embedding and generation-service failures were diagnosed separately.
