
Ki-Ung Song
Problem Solver with AI and Mathematics Who Wants to Change the World
Seoul, South Korea
Featured Updates
08-09-2024
Neural Solver Towards Future of Simulation: Deep DivePost
08-09-2024
Neural Solver Towards Future of Simulation: ExplorationPost
Since May 2023
LLM4FinanceProject
Tags
See All >>LLM4Finance
May 2023 - Current
Developing an LLM-powered finance dashboard to optimize personal investment
strategies, incorporating financial engineering techniques.
Pipeline Outline

MLOps Pipeline for Sentiment Analysis
- PEFT-based LLM predicts news sentiment: positive, neutral, or negative.
- Continuous training triggers in Hugging Face Spaces once new labeled data exceeds a threshold.
- Continuous deployment in a separate Hugging Face Space.
- Model quantized via llama.cpp for CPU-based deployment.
- Hugging Face Model Hub used for saving/loading trained models.
LLM Application
- LLM agents automate data processing: sentiment labeling and quality control.
- Addressed performance issues from data quality errors (punctuation, duplication).
- LLM assistant streamlines financial statement analysis.
- Experimental: RAG-based framework for enhanced analysis.
- Experimental: Multi-agent LLM simulates investment environments tailored to different investor personas.
Featured Updates
08-09-2024
Neural Solver Towards Future of Simulation: Deep DivePost
08-09-2024
Neural Solver Towards Future of Simulation: ExplorationPost
Since May 2023
LLM4FinanceProject