CS undergraduate at National Taiwan University. Software engineer focused on LLM systems and AI infrastructure, with applied research in quantitative trading.
Working paper on LLM cryptocurrency trading evaluation — embargo-period 2×2+1 prompt-ablation on ETH/USDT with multi-seed replication, comparing Random Forest, Claude Sonnet 4, and a QLoRA-fine-tuned Qwen3-14B agent. The RF arm is deployed in a live execution layer (regime-filter wrapper) running with real capital. Pre-submission.
Secondhand price-tracking platform for PVC collectible figures in Taiwan. Real transaction data, trend analysis, and a Chrome extension for in-browser reporting.
Delta-neutral funding-rate arb on USDT-M perpetuals: shorts spot (cross-margin borrow) against a long perp leg (or the reverse). Two modes — carry (hours to days, sustained funding) and snipe (~2 min, extreme funding spikes near settlement) — with dual-margin monitoring sized to survive ±30% price moves on either leg independently.
Parses unstructured Telegram trading-channel signals with a local Ollama LLM and forwards structured orders to Binance Futures — zero cloud, fully on-device inference. The project that pulled me into quantitative trading.
On-prem RAG knowledge base built on DeepSeek and vLLM. No data leaves the local network; OpenAI-compatible endpoints for drop-in tooling. Deployed on HPC infrastructure.
Junior in Computer Science & Information Engineering at National Taiwan University. Primarily a software engineer focused on LLM systems and AI infrastructure — building things that have to work, not just benchmark well.
My research interest is evaluation honesty: how to tell whether an AI system is actually doing the thing you think it is. Quantitative trading is the testbed I work in most — markets give live, falsifiable feedback that a benchmark can't.
Currently looking ahead at graduate study and roles in LLM systems, AI infrastructure, or applied research. Open to collaboration.