Posts
All the articles I've posted.
-
FEMS Project #2 — A Real 230-Page Corpus + Chunk Quality Gates + the Chinese-Character Mixing Trap in a Local LLM
Built a corpus of roughly 230 pages / 35,000 rows from the Korea BEMS Association guides, Korea Energy Agency materials, and the UCI Steel dataset. After paragraph-based chunking (target 800 chars) + bge-m3 + Chroma indexing, a quality gate (ratio of complete Hangul/ASCII characters) excluded 7 chunks from table-of-contents pages. Then I hit a trap — qwen2.5:7b mixed in Chinese characters on the second question and suffered generation collapse (spitting out unrelated Chinese city coordinates as GeoJSON). Partially fixed with temperature / system prompt → ultimately switched to exaone3.5:7.8b for clean handling.
-
LangGraph Study #3 — Human-in-the-Loop (Approving Risky Tools) · Checkpoints and the Interrupt Mechanism
When an agent automatically calls risky tools like file deletion, sending emails, or payments, things go wrong. Implementing the HITL pattern using LangGraph's checkpoint + Interrupt mechanism to get human approval right before a risky tool call. An unexpected trap: the LLM's own safety guard can block execution before it ever reaches HITL. How to separate the system prompt from the HITL gate.
-
FEMS Project #1 — Comparing Low-Spec Local Setup (Ollama + bge-m3 + Chroma) vs Claude API RAG
Building a RAG prototype while studying the FEMS (Factory Energy Management System) domain. Comparing local LLM inference (Ollama) on a low-spec environment (GTX1660 Super, 6GB VRAM) against calling the Claude / OpenAI APIs. Using bge-m3 for embeddings (strong Korean support) and Chroma as the vector DB. Ollama's cold start of 95 seconds dropped to 10 seconds after warm-up, with accuracy matching the cloud.
-
Quant Study 00 — pandas Basics · Trend Following vs Mean Reversion · The 4 Backtest Biases · Fake Alpha
Starting a quant learning series. pandas Series/DataFrame, loc/iloc, the warm-up period (NaN), silent data corruption from adjusted close prices, trend following (MA, golden/dead cross) vs mean reversion, the 4 backtest biases (look-ahead, overfitting, unrealistic trading costs, survivorship bias), and why you need to look at a distribution instead of a single point to filter out fake alpha (B&H beat rate / median alpha / market exposure / walk-forward).