One-on-one research advising (advisor: Yang Yangrui). After building programming and modelling foundations through contests, shifted focus to research and academic writing: literature search & close reading on RAG, knowledge graphs and retrieval reranking; independently completed a full empirical study (topic → design → analysis → writing). Studied the SciToolAgent paper (Nature Computational Science, 2025) and presented it to the group.
AI undergrad @ NCWU · open to research & collaboration
Hi, I'm Raelon Veritas Lee
AI Undergrad — Retrieval / RAG · Machine Learning · Applied Research
I turn research into systems that ship — from a filed patent on intelligent bid-document analysis to empirical retrieval studies and deep-learning signal separation. Scroll for patents, papers and projects.
~/portfolio.json
- projects
- 14
- patents
- 01
- papers
- 1+
- stack
- Python · PyTorch · RAG
- location
- China
- status
- open to work
01 selected work
Projects
- context-retrieval★ Featured
Contextual Retrieval: Empirical Study
Reproduced Anthropic's Contextual Retrieval, then ran systematic comparisons across domains — finding a 'double-edged sword' effect and the limits of knowledge graphs.
- Python
- RAG
- Contextual Embeddings
- BM25
- Knowledge Graph
3 phases · 6 controlled experiments · reproducible
- gcms-separation★ Featured
GC-MS Signal Deconvolution & Identification Engine
End-to-end 1D-CNN for GC-MS: identify components of essential-oil mixtures and predict blend ratios directly from 1D chromatographic signals. Enterprise exploration.
- Python
- 1D-CNN
- PyTorch
- GC-MS
- Synthetic Data
87 monomers · 74,820 samples · ~99% blind-test accuracy (synthetic)
- bid-analysis★ Featured
Intelligent Bid-Document Analysis System
University–enterprise project: intelligent parsing and fast screening of bidding documents. Core developer through requirements, architecture and iteration — results filed as an invention patent (published).
- Python
- LLM
- RAG
- FastAPI
Invention patent published · enterprise software engineering
🔒 Private · demo on request Details →
02 patents & papers
Research
- paperPaper In preparation (unpublished)
A Systematic Comparison of Retrieval-Augmentation Methods for Chinese Vertical Domains
First-author empirical study reproducing Anthropic's Contextual Retrieval and comparing four routes (hybrid baseline, CR, knowledge graph, cross-encoder reranking) on real flood-prevention plans.
2×2 ablation · n=30 significance test · code open-sourced
- patentPatent Published · under substantive examination
Bid-Document Intelligent Analysis & Recommendation Method, Device and Storage Medium
Invention patent (published, under substantive examination) — the core method behind the bid-analysis system. Applicant: NCWU; one of the inventors.
申请号 202610250247.1 · 公布于 2026-06-09（第 42 卷 2401 期）
03 timeline
Experience
- Sep2025 - NowResearch Training · Advisor's Group (NCWU)Undergraduate Researcher — Retrieval / RAG
- Mar2025 - Mar2026University–Enterprise Project · Bid-Analysis SystemCore Developer
Core developer on my advisor’s university–enterprise collaboration: requirement analysis, architecture design and iterative development of an intelligent bid-document analysis system. Gained enterprise-grade software-engineering and vertical-domain AI experience; results filed as an invention patent (published).
- Dec2024 - Jan2025University–Enterprise GIS Training · Wuhan Zondy CyberTrainee (graduated with distinction)
University–enterprise GIS development training at Wuhan Zondy Cyber (中地数码). Built a smart-city GIS visualization prototype (Vue 3 + Leaflet + AntV + Three.js), completed the required coursework, graduated with distinction. Public edition open-sourced as NCWU-SmartCity-WuHan.
04 honors & contests
Awards & Competitions
- Provincial 🏆 Provincial First PrizeProblem C · team of 3 · Henan Region · CSIAM
- Provincial 🏆 Provincial Third Prize16th Lanqiao Cup — National Software & IT Talent ContestC/C++ Programming · University Group B · Henan Region · MIIT Talent Center
- School 🏆 Second PrizeFreshman Career-Planning Essay Contest (2024)NCWU Career & Innovation Center
Let's build something worth building
Want to collaborate, discuss an opportunity, or just say hi — reach me by email or social.