About

Hyeonseop Yoon (윤현섭)

I build grounded, evidence-driven language systems and connect retrieval research with production constraints.

Through-line

I started in language and cognitive-neuroscience research, comparing how brains and NLP models represent meaning. Today I build applied NLP systems around retrieval, grounding, evaluation, model adaptation, and serving.

The connection is methodological: define what a result means, test the claim on an appropriate split, expose limitations, and give the system an abstention path when evidence is insufficient.

Capabilities

retrieval and reranking evaluationgrounded QA and abstention designembedding fine-tuning and model cardsLLM adaptation and servingPyTorch, Hugging Face, vLLMDocker, Postgres/pgvector, SLURM/H100

Links

GitHub Portfolio repository Hugging Face ORCID Email

Public/private boundary

Public materials contain only shareable methods, code, models, and evaluation context. Client identities, production repositories, raw logs, internal infrastructure, and private study or diary content are excluded.