PinnedAkshat A. Mistry·Sep 24Byte-Pair Encoding (BPE) TokenizersAn end-to-end practical guide to how modern LLM tokenizers work
PinnedAkshat A. Mistry·Jul 15Part IV | Multi-Agent Coordination: The Hub-and-Spoke Architecture“Sāṅkhya” CCA-F Study Series · Part 4 · Domain 1 — Agentic Architecture & Orchestration
PinnedAkshat A. Mistry·Jul 14Part III | ‘tool_use’, End to End: The Agentic Loop in Three IterationsAnother post in our “Sāṅkhya” CCA-F Study Series, Part 3 of the road to Claude Certified Architect — Foundations
PinnedAkshat A. Mistry·Jul 12Claude 101: The Foundation to CCA-FOne of the first posts under “Sāṅkhya” CCA-F Study Series · Part 1 of the road to Claude Certified Architect — FoundationsA response icon1A response icon1
PinnedAkshat A. Mistry·Jul 13Part-II | The Seven ‘stop_reason’ ValuesAnother post under “Sāṅkhya” CCA-F Study Series · Part 2 of the road to Claude Certified Architect — Foundations
Akshat A. Mistry·Oct 30Multi-Agent Systems and Architectural Value; CCAR-P Domain-1A multi-agent architecture is justified only when a single-agent topology cannot satisfy the project requirements, and its adoption…A response icon1A response icon1
Akshat A. Mistry·Oct 28Choosing the Right Pattern: Design Decisions to Defend on the CCAR-P ExamPattern choice is the first architectural decision, and the hardest to reverse. Pick the simplest pattern that meets the requirement, and…A response icon1A response icon1
Akshat A. Mistry·Oct 16Eigenvectors and EigenvaluesEigenvectors are the special vectors that keep their original direction under the transformation.
Akshat A. Mistry·Oct 3Token IDs Are Labels: Why One-Hot Encoding Removes False GeometryToken IDs are vocabulary labels, not measurements. This article explains why raw integer IDs and scalar transformations create false order…
Akshat A. Mistry·Aug 9ELMo, Functionally ExplainedMachine Learning Foundations · The character-level pass, the word-level biLSTM pass, the word2vec substitution point, and what happens…