Series
Multi-part deep dives — each series builds one thing end to end, with experiments you can rerun yourself.
AI Memory Mastery
1 partComplete guide to building a permanent personal knowledge system with Obsidian + Claude Code. Wiki + Memory dual-axis architecture.
Start from Part 1 →LLM Inference
4 partsBuild Self-Attention from scratch. Compare MHA → GQA → MQA evolution in code. KV Cache mechanics and Prefill vs Decode analysis.
Start from Part 1 →Lora Finetuning
3 partsFine-tune a 7B model on a T4 16GB with QLoRA. Dataset construction, training execution, Wandb monitoring, and Before/After comparison.
Start from Part 1 →AgentScope
5 partsBuild multi-agent systems with SequentialPipeline, FanoutPipeline, and MsgHub. Practical code review team pattern.
Start from Part 1 →AutoResearch
3 partsFrom environment setup to agent execution and overnight results analysis. Tuning guide for smaller GPUs and practical tips.
Start from Part 1 →DeerFlow Multi Agent
4 partsDeerFlow 2.0 architecture, setup, and first task execution. A SuperAgent runtime with 9 agent nodes, 5 tool sources, and Docker sandboxes.
Start from Part 1 →Qwen3.5 Vs DeepSeek V3.2
2 partsStep-by-step guide to running Qwen 3.5 locally. From 5-minute Ollama setup to production vLLM servers, plus optimal model size selection per GPU.
Start from Part 1 →Agentic RAG
2 partsImplement Self-RAG reflection tokens and CRAG quality-based fallback. Build retry/fallback logic with LangGraph conditional edges.
Start from Part 1 →LLM Agent
3 partsUpgrade ReAct with Tool Calling, then build Reflection and Planning Agents with LangGraph.
Start from Part 1 →Backprop From Scratch
1 partHow microgpt.py's 15-line backward() works. From high school calculus to chain rule, computation graphs, topological sort, and backpropagation.
Start from Part 1 →microGPT
1 partA line-by-line dissection of microgpt.py -- a pure Python GPT implementation with zero dependencies. Training, inference, and autograd in 150 lines.
Start from Part 1 →Diffusion LLM
3 partsD3PM, Transition Matrices, Absorbing States, MDLM -- how to bring diffusion from continuous space to discrete tokens.
Start from Part 1 →LLM Reasoning Failures
3 partsReversal Curse, Counting, Compositional Reasoning — fundamental Transformer failures tested across 7 models.
Start from Part 1 →Logit Lens To Tuned Lens
2 partsUsing TransformerLens to directly manipulate model activations, we trace which layers and heads causally produce the answer. A hands-on guide to activation patching.
Start from Part 1 →Demo To Prod Part1
3 parts5 Non-Negotiable Standards for Enterprise Deployment
Start from Part 1 →Advanced RAG
7 partsMany teams try to solve RAG accuracy problems by tuning their vector database. But the real bottleneck is chunking that ignores the relational structure of source data.
Start from Part 1 →Ontology Knowledge Graph RAG
1 partVector search alone isn't enough. Upgrade your RAG system with Knowledge Graphs that understand entity relationships.
Start from Part 1 →Claude Code Context
3 partsWhat if Claude automatically ran lint, tests, and security scans every time it generated code? Learn how to automate team workflows with Hooks.
Start from Part 1 →Claude Code MCP
1 partWhat if Claude could read Jira tickets, send Slack messages, and query your database? Learn how to extend Claude's capabilities with MCP servers.
Start from Part 1 →SQL Analytics
11 partsRow Explosion in 1:N JOINs and how to aggregate revenue correctly
Start from Part 1 →DDPM Diffusion Models
10 partsHow Latent Space solved the memory explosion problem of pixel-space diffusion. Complete analysis from VAE compression to Stable Diffusion architecture.
Start from Part 1 →