Blog
Writing & updates
Daily education essays for students and independent creators.
Building a Local Multi-Agent Workflow with CrewAI
A practical walkthrough of Agent, Task, Crew, Process.sequential, and deterministic post-processing validation using CrewAI with a local Ollama model.
Understanding LlamaIndex Architecture Through a RAG Application
A practical tour of how Document, TextNode, VectorStoreIndex, StorageContext, PGVectorStore, and QueryEngine work together in a Chinese movie recommendation RAG demo.
Understanding How Models Learn to Prefer Better Answers Through a DPO Alignment Demo
A practical DPO Alignment walkthrough that explains where Alignment fits in the LLM generation pipeline, how DPO data and training roles work, and how DPO differs from SFT.
LangGraph Workflow Basics: Understanding Node, State, and Edge Through a Story Demo
A practical walkthrough of a LangGraph story demo, explaining Node, State, and Edge and showing how AI workflows can support branching, loops, review, retries, and stop conditions.
Supervised Fine-Tuning: From SFT to LoRA Practice
A practical walkthrough of supervised fine-tuning with Llama 3.1, Colab, Unsloth, LoRA adapters, Alpaca-style data, checkpoints, and before/after evaluation.
Modern Agent Architecture Explained
A practical explanation of modern agent architecture: model, tools, memory, planning, execution, governance, and a travel-planning agent example.
Introducing MCP Proxy to an API Call Demo
Add an extensible MCP tool-calling path to an existing client-server API demo without breaking the original structure.
From Chatting to Getting Things Done: AI Skills and Google Workspace
How reusable AI Skills turn instructions, tools and boundaries into dependable workflows across Google Workspace.
The Technical Architecture of a Personal Website
A right-sized blueprint for containers, reverse proxies, persistent data, backups and a reversible delivery pipeline.
Telegram vs. Other Social and Messaging Platforms
From cloud sync, privacy models and large communities to bots, Mini Apps and AI agents: a practical comparison of Telegram, WhatsApp, Signal and Discord.
The Industry Baseline for LLM Training Data Scale
From hundreds of billions to trillions of tokens: a practical map of LLM training-data scale, public facts, and industry estimates.
Skill Architecture Through a File Analysis Skill
A file-analysis skill example explaining how modern agents use metadata, SKILL.md, references, and scripts through progressive disclosure.