Hyosang K. - Student Assistant

Berlin, Germany

Experience

Sep 2024 - Aug 2025

1 year

Enhanced in-car voice assistant responsiveness by optimizing engineering solutions for audio data processing.
Improved dataset quality and model robustness by generating synthetic data with Hugging Face sources.
Increased contextual relevance of voice assistant outputs through optimized AI prompt engineering.
Reduced transcription errors by debugging inconsistencies between audio inputs and transcripts.
Boosted team productivity and software delivery speed by automating workflows and actively driving Agile updates.
Mapped intents across different datasets; standardized data using JSON.
Created annotation guidelines and evaluated annotator quality.
Generated diverse synthetic utterances using LLMs, including Korean.
Used clustering to assign utterances to intents and find sub-intents.
Identified and removed noisy or duplicate utterances using LLMs.
Created visualizations to analyze data and improve dataset quality.

Mar 2023 - Aug 2024

1 year 6 months

Enabled advanced analysis of large-scale datasets by engineering text clustering solution with BERT, distilBERT, RoBERTa, and text-embedding-ada-002.
Tackled information trust issues in crisis scenarios by developing search engine modules with Elasticsearch on Docker.
Expanded research data sources by automating API-call scripts to retrieve YouTube transcripts and comments.
Facilitated research collaboration by leading virtual workshops with structured discussions and participant voting.
Conducted scientific literature searches.
Performed data maintenance and processing with Excel and Python data pipelines, reporting and presenting results.
Implemented analysis of text clustering methods using BERT-based transformers.
Developed prototype of search engine for a COVID-19 social media data knowledge base using Python, Docker, Jupyter, and ElasticSearch.
Implemented Python scripts for data retrieval from YouTube and preprint servers.
Supported internal documentation of technical procedures, such as guideline for GPU access in the department.
Moderated and supported the organization of a SciBeh’s virtual workshop 2024 on the topic Epistemic Trespassing.