AI / ML

• Claude Code — Best Practices

  How to organize work with Claude Code: project instructions, rules and skills, a task and bug tracker in the repository, planning before code, subagents, context, permissions and budget.

• Claude Code — commands, settings, flags

  A Claude Code reference: every slash command with its arguments broken down and examples, settings.json keys, permission rules, terminal subcommands, flags and environment variables. Current as of August 2026.

• Claude Code Hooks

  A walkthrough of Claude Code hooks: every event, exit codes and JSON decisions, matcher and if, the five handler types, and three working hooks — sound, subscription limits and a work log.

• LangChain and LangGraph — agents, graphs and the infrastructure around them

  How the LangChain 1.x and LangGraph 1.x stack fits together: models and messages, tools, create_agent and middleware, StateGraph with reducers, checkpointers and store, streaming, human-in-the-loop, LangSmith, Agent Server and the langgraph CLI, plus the 0.x migration.

• LLama.cpp AI LLM Engine — How It Works

  How inference works inside llama.cpp: loading a GGUF model, tokenization, building the compute graph, the KV cache, and token sampling.

• LoRA — Final training of AI LLM models

  Fine-tuning an LLM with LoRA: when it beats RAG, how to train adapters for gemma-3-4b with unsloth and peft, and export the result to GGUF.

• Neural networks in simple terms

  How an LLM actually works: tokenization, embeddings, transformer attention, quantization, and the samplers that pick each next token.

• Offline AI Launcher — launch AI LLM neural networks on smartphone

  Offline AI Launcher runs LLMs like Gemma, DeepSeek, Llama and Qwen straight on your Android phone from .gguf files, with no internet or subscription.

• Prompt engineering techniques for Large Language Models

  The effort knob on reasoning models, strict output schemas, agent prompts, delimiters, XML tags and few-shot — how to write prompts for ChatGPT, Claude or Gemini and get predictable output.

• RAG (Retrieval-Augmented Generation), Function / Tools Calling, CAG (Cache-Augmented Generation) for LLM models

  RAG, function calling and CAG for local LLMs: how vector search works, with working examples in Kotlin, Dspy, LangChain, LangChain4j and llama.cpp.

• Vector databases

  How vector databases work: embeddings, HNSW indexing and similarity search, plus a comparison of Milvus, Qdrant and Weaviate with Milvus code.