A working map of organizational knowledge and its tacit layer: how we define it, why it resists being written down, and how structured elicitation tries to bring it out.

Why dense retrieval alone misses relationship-heavy queries, and how combining it with a knowledge graph fixes part of the gap.

RAG quality is a causal chain, not one score. A detailed guide to retrieval, generation, citation, robustness, judge reliability, and task-level evaluation.

How ReAct, structured tool calling, and MCP fit together: a paper-grounded look at agent loops, tool interfaces, and the gap between valid calls and reliable outcomes.

How we taught IndoBERT to tag Indonesian financial regulations with several compliance labels at once, using augmentation-free contrastive pre-training and semi-supervised pseudo-labeling.