Hello!
I am Minji Kim, an aspiring AI researcher and engineer focused on building practical systems grounded in solid research.
My recent work centers on multimodal and LLM-based methods that bridge academic research with real-world applications.
SKILLS
AI Research & Multimodal Systems
Grounded Multimodal NERMultimodal Information ExtractionVision-Language Model Fine-TuningKnowledge Graph Question Answering (KGQA)Temporal Knowledge Graph Forecasting
Applied LLM & Data-Centric AI
Software Engineering & MLOps
RESEARCH
Grounded Multimodal Named Entity Recognition (GMNER)
2025.03 – 2026.02 (Paper Under Review)- Proposed an anchor-guided collaborative grounding framework for grounded multimodal NER.
- Addressed ambiguities in visually similar entities by modeling inter-entity relationships and fine-grained entity types.
- Achieved state-of-the-art performance on GMNER benchmarks, outperforming large-scale pretrained vision–language models in grounding precision.
GMNERMultimodal Information ExtractionVision-Language ModelEntity Grounding
Knowledge Graph Question Answering (KGQA)
2024.03 – 2025.02- Proposed a novel multi-hop KGQA model that explicitly models question intent.
- Identified relation-specific keyphrases and predicted answer entity types using KG schema information.
- Achieved superior performance on WebQSP and CWQ compared to state-of-the-art models.
Knowledge GraphKGQATransformersIntent Modeling
Protein–Peptide Docking Complex Prediction
Mar 2023 – Feb 2024- Studied a GNN-based decoy evaluation model for protein–peptide docking complex prediction.
- Analyzed residue-level interactions to assess docking accuracy and interaction quality.
- Participated in the “Deep Learning–Based Ultra-Fast Virtual Peptide Screening.” funded by the National Research Foundation of Korea
Graph Neural NetworkBioinformaticsProtein DockingDeep Learning
INDUSTRY EXPERIENCE
Place AI Local Domain Classification Development
NAVER Corporation · Internship · Dec 2025 – Jan 2026
- Designed and implemented a CLI tool for test data versioning and archiving to support large-scale evaluation workflows.
- Constructed IR test datasets using rerankers and LLM-as-a-Judge evaluation frameworks.
- Designed a local-domain query detection pipeline, including data construction, taxonomy definition, and LLM-based labeling.
- Trained and evaluated encoder-based models, achieving 93.8% accuracy and 260 QPS.
- Conducted supervised fine-tuning experiments on small language models (SLMs).
Information RetrievalLLM EvaluationData ConstructionSLM Supervised Fine-TuningEncoder Fine-Tuning
ENGINEERING
Personal Portfolio Website
- Designed and developed a personal portfolio website using Next.js.
- Implemented reusable UI components and deployed the website.
Next.jsTypeScriptTailwind CSS
Performance Profiling and Optimization of Inference Pipeline
- Profiled latency bottlenecks in an LLM inference pipeline using PyTorch profiler.
- Reduced end-to-end inference latency by adopting FP16 inference and mixed-precision optimization.
PyTorchInference OptimizationFP16 InferenceMixed PrecisionPerformance Profiling
Fine-tuning Vision-Language Models for GMNER
- Fine-tuned an open-source vision–language model (LLaVA-v1.5-7B) for entity grounding in grounded multimodal named entity recognition (GMNER).
- Explored the feasibility of end-to-end entity grounding using a VLM when entity spans and types are provided as structured inputs.
- Designed prompt templates and data formatting strategies tailored to GMNER tasks.
- Conducted experiments on the Twitter-GMNER and Twitter-FMNERG datasets to evaluate grounding performance and inference latency.
PythonPyTorchVision-Language ModelLLaVAPrompt EngineeringMultimodal Fine-Tuning
HANDY Web Platform Development
- Developed the complete backend for an AI-based sign language learning web service aimed at lowering entry barriers for Korean Sign Language education.
- Designed and implemented RESTful APIs for user authentication, dictionary search, vocabulary management, and learning progress tracking.
- Built database schemas and backend logic to support multiple user-defined vocabulary lists and learning features.
- Deployed and operated the backend services on AWS and Cloudtype, ensuring stable online availability.
Node.jsMySQLREST APIPostmanAWSCloudtypeBackend Development
🏆 Excellence Award (Third Place), Graduation Exhibition, Dept. of Computer Engineering, Hongik University
PUBLICATION
[1] Anchor-Guided Collaborative Grounding for Context-aware Grounded Multimodal Named Entity Recognition, Under Review Minji Kim, Yusol Oh, Midan Shim, Kaehyun Um, Kyong-Ho Lee
PROFILE
Name
Minji Kim
Birth
1999.12
Location
Seodaemun-gu, Seoul
Phone
+82 10-5590-1082
Email
mjluckk@gmail.com
Education
Hongik University - Computer Engineering (B.E., 2019-2024)
Yonsei University - Artificial Intelligence (M.S. 2024-Present)