https://relu.chat/blog/ Technical articles about on-device AI, browser-based chatbots, NLP, and privacy-first machine learning. en Wed, 09 Sep 2026 14:41:29 GMT https://relu.chat/assets/logo.png https://relu.chat https://relu.chat/blog/dot-product-cosine-worked-example/ https://relu.chat/blog/dot-product-cosine-worked-example/ Wed, 09 Sep 2026 14:00:00 GMT Work through a dot product, vector norms, and cosine similarity, including the zero-vector case that retrieval code must handle. Yunus Emre Vurgun (https://yunusemrevurgun.com) linear-algebra worked-example retrieval https://relu.chat/blog/grouped-validation-leakage/ https://relu.chat/blog/grouped-validation-leakage/ Wed, 09 Sep 2026 14:00:00 GMT Why random row splits overstate performance on repeated users, devices, or experiments, and how grouped validation changes the question. Yunus Emre Vurgun (https://yunusemrevurgun.com) data-science evaluation https://relu.chat/blog/least-squares-without-inverse/ https://relu.chat/blog/least-squares-without-inverse/ Wed, 09 Sep 2026 14:00:00 GMT Derive a two-parameter fit, check its residuals, and understand why practical solvers use QR or SVD instead of an explicit inverse. Yunus Emre Vurgun (https://yunusemrevurgun.com) linear-algebra data-science worked-example https://relu.chat/blog/local-chat-privacy-network/ https://relu.chat/blog/local-chat-privacy-network/ Wed, 09 Sep 2026 14:00:00 GMT Distinguish local question processing from page downloads, browser storage, external links, and offline availability. Yunus Emre Vurgun (https://yunusemrevurgun.com) privacy browser-ai web-platform https://relu.chat/blog/nash-equilibrium-best-response-table/ https://relu.chat/blog/nash-equilibrium-best-response-table/ Wed, 09 Sep 2026 14:00:00 GMT Check a complete 2×2 game, identify two pure equilibria, and distinguish unilateral stability from efficiency. Yunus Emre Vurgun (https://yunusemrevurgun.com) game-theory worked-example https://relu.chat/blog/precision-recall-small-example/ https://relu.chat/blog/precision-recall-small-example/ Wed, 09 Sep 2026 14:00:00 GMT Compute precision, recall, accuracy, and F1 from the same small confusion matrix, then connect thresholds to the errors you care about. Yunus Emre Vurgun (https://yunusemrevurgun.com) data-science worked-example https://relu.chat/blog/q-learning-sarsa-worked-example/ https://relu.chat/blog/q-learning-sarsa-worked-example/ Wed, 09 Sep 2026 14:00:00 GMT Calculate Q-learning and SARSA updates on the same transition, and see why their targets differ during exploration. Yunus Emre Vurgun (https://yunusemrevurgun.com) reinforcement-learning worked-example https://relu.chat/blog/rank-svd-conditioning/ https://relu.chat/blog/rank-svd-conditioning/ Wed, 09 Sep 2026 14:00:00 GMT Use a diagonal matrix to distinguish exact rank, low-rank approximation, and sensitivity to numerical perturbations. Yunus Emre Vurgun (https://yunusemrevurgun.com) linear-algebra numerical-computing https://relu.chat/blog/reading-islamic-science-sources/ https://relu.chat/blog/reading-islamic-science-sources/ Wed, 09 Sep 2026 14:00:00 GMT Distinguish a scholar’s work, surviving manuscripts, later biographies, and modern interpretations without reducing history to invention lists. Yunus Emre Vurgun (https://yunusemrevurgun.com) history-of-science sources https://relu.chat/blog/reward-design-small-chatbot/ https://relu.chat/blog/reward-design-small-chatbot/ Wed, 09 Sep 2026 14:00:00 GMT Design routing rewards without confusing longer answers, topic overlap, or training scores with useful explanations. Yunus Emre Vurgun (https://yunusemrevurgun.com) reinforcement-learning evaluation chatbots https://relu.chat/blog/rl-termination-truncation/ https://relu.chat/blog/rl-termination-truncation/ Wed, 09 Sep 2026 14:00:00 GMT A small numerical example shows why environment termination and time-limit truncation should not share one done flag. Yunus Emre Vurgun (https://yunusemrevurgun.com) reinforcement-learning debugging https://relu.chat/blog/service-worker-model-updates/ https://relu.chat/blog/service-worker-model-updates/ Wed, 09 Sep 2026 14:00:00 GMT Separate stable embedding assets from mutable policies, avoid eager runtime downloads, and test update and offline failure paths. Yunus Emre Vurgun (https://yunusemrevurgun.com) web-platform performance offline https://relu.chat/blog/tiny-policy-training-cpu/ https://relu.chat/blog/tiny-policy-training-cpu/ Wed, 09 Sep 2026 14:00:00 GMT A reproducible CPU training recipe, a real three-second run, and the limits of routing metrics on a small contextual-bandit policy. Yunus Emre Vurgun (https://yunusemrevurgun.com) reinforcement-learning engineering evaluation https://relu.chat/blog/wasm-javascript-policy-parity/ https://relu.chat/blog/wasm-javascript-policy-parity/ Wed, 09 Sep 2026 14:00:00 GMT A small MLP shows why export parity, fixed memory, startup costs, and end-to-end latency matter more than assuming WASM is faster. Yunus Emre Vurgun (https://yunusemrevurgun.com) webassembly performance engineering https://relu.chat/blog/retrieval-metrics-precision-recall-mrr/ https://relu.chat/blog/retrieval-metrics-precision-recall-mrr/ Thu, 06 Aug 2026 17:30:00 GMT How to evaluate a knowledge-base retriever with precision, recall, MRR, and NDCG: definitions, worked examples, and a tuning loop for dense/sparse ensembles. Yunus Emre Vurgun (https://yunusemrevurgun.com) retrieval ndcg mrr evaluation https://relu.chat/blog/chunking-strategies-knowledge-base/ https://relu.chat/blog/chunking-strategies-knowledge-base/ Thu, 06 Aug 2026 16:45:00 GMT How chunk size, boundaries, and overlap decide what a knowledge base retriever can find — fixed, overlapping, and structure-aware chunking compared. Yunus Emre Vurgun (https://yunusemrevurgun.com) chunking retrieval knowledge-base embeddings bm25 https://relu.chat/blog/query-expansion-without-llm/ https://relu.chat/blog/query-expansion-without-llm/ Thu, 06 Aug 2026 15:30:00 GMT Improve retrieval without an LLM: index-time alias weighting, query-side synonym expansion, edit-distance spelling correction, and boosting before re-ranking. Yunus Emre Vurgun (https://yunusemrevurgun.com) query-expansion bm25 spelling-correction retrieval https://relu.chat/blog/knowledge-base-fragments-fields-metadata/ https://relu.chat/blog/knowledge-base-fragments-fields-metadata/ Thu, 06 Aug 2026 14:00:00 GMT Knowledge base entries built from categorized fragments, weighted fields, and confidence metadata — and why structure decides retrieval quality. Yunus Emre Vurgun (https://yunusemrevurgun.com) knowledge-base retrieval metadata bm25 rag https://relu.chat/blog/indexeddb-client-storage/ https://relu.chat/blog/indexeddb-client-storage/ Thu, 06 Aug 2026 13:00:00 GMT Store embeddings as binary and session state as records with IndexedDB: schema design, vector search, transactions, and local-first privacy. Yunus Emre Vurgun (https://yunusemrevurgun.com) indexeddb storage embeddings local-first privacy https://relu.chat/blog/float32-vs-float64-precision/ https://relu.chat/blog/float32-vs-float64-precision/ Thu, 06 Aug 2026 12:30:00 GMT When to use float32 versus float64 in browser ML: tensors stay float32, accumulators like EMA summaries deserve float64, and int8 quantization shrinks weights. Yunus Emre Vurgun (https://yunusemrevurgun.com) float32 float64 precision quantization