Practical recipe: session embeddings + HDBSCAN + LLM-named clusters for chat topics
leland_mcinnes · x · 2026-09-13
- A reply lays out two reproducible approaches for topic-classifying chat sessions: embed sessions (or their summaries), cluster with HDBSCAN (optionally UMAP dimensionality reduction first), and have an LLM name each cluster with descriptive analysis.
- If you already have fixed categories, run an agent at the end of each conversation to classify the session into strict JSON (or classify incrementally), and for semantic search store the same embeddings in a vector store with metadata filtering before kNN.
More from coding & agent
- Agents get stronger weekly yet still solve the wrong problem: AI takeover ETA? — BarriosA2I · 2026-09-13
- Swarms revamps chat UI with agent controls and 2,000+ models via its API — KyeGomezB · 2026-09-13
- "Nobody wrote the Matrix" essay lands on Reddit, arguing AI coding dies by 2036 — TMWNN · 2026-09-13
- After millions of tokens: Muse Spark 1.3 High matches Sol Medium at a fraction of the price — HumungreousNobolatis · 2026-09-13
- RoastMyHarness: open-source tool benchmarks your custom Pi agent setup against bare Pi on DeepSWE — AnotherObsceneBean · 2026-09-13
- One question to ask your agent after every session: 'which assumption, if wrong, changes your answer most?' — andrew_n_carr · 2026-09-13