Wiki Operation Log

Maintained automatically by AI. Reverse chronological order.

2026-04-09 (Project Deep Dives)

new DeerFlow SuperAgent Framework — ByteDance’s LangGraph SuperAgent with Docker sandbox, 5 agent roles
new Mem0 Memory Architecture — LLM extraction pipeline, graph memory, Mem0 vs Letta decision framework
new Mastra Agent Framework — TypeScript-native with Observational Memory (94.87% LongMemEval)
update Claude Code Architecture Deep Dive — Added 5 missing sections: hidden features (88 flags), KAIROS daemon, Buddy system, anti-distillation, Undercover Mode

new AI Product Architecture — End-to-end system design: auth, billing, API, frontend, queues, multi-tenant
new AI UX Patterns — Streaming UI, loading states, error handling, feedback, conversation UX
new AI Product Metrics — Business metrics: task completion, deflection, CSAT, cost per conversation
new LLM Observability and Monitoring — Production monitoring, tracing, OpenTelemetry, debugging, cost alerts
new Data Engineering for AI — RAG pipelines, chunking, embeddings, data quality, synthetic data, PII

new LLM Security and Guardrails — OWASP Top 10, prompt injection, guardrail frameworks, layered defense
new Production Prompt Engineering — System prompt architecture, caching economics, extended thinking, meta-prompting
new Fine-tuning vs Prompting vs RAG — Decision framework, LoRA/QLoRA, distillation patterns
new LLM Inference and Serving — vLLM, quantization methods, hardware guide, cloud providers
new AI Cost Optimization — Prompt caching, model tiering, batch API, production cost examples

new MCP Architecture and Ecosystem — Protocol spec, transports, OAuth, adoption, 1000+ server ecosystem
new AI Coding Tools Landscape 2026 — Full competitive map: Claude Code, Cursor, Copilot, Codex, Devin, Kiro, Aider
new AI Evals and Testing — Eval platforms, EDD workflow, RAG testing, agent evaluation patterns

new Claude Code Architecture Deep Dive — Deep analysis from source leak: agent loop, memory, subagent models, 10 commercial patterns
new Open Source LLM Landscape 2026 — Gemma 4, Qwen 3.5/3.6, Llama 4, DeepSeek comparison with benchmarks
new Multi-Agent Architecture Patterns — Production frameworks + Claude Code’s 3 subagent models
new Agent Memory Systems — Claude Code 7-layer + Letta/MemGPT + Mem0 + autoDream analysis
new RAG Architecture Patterns 2026 — Hybrid search, GraphRAG, Agentic RAG, contextual retrieval
new Karpathy LLM Education Stack — Merged MicroGPT + nanoGPT + llm.c + Autoresearch into one page
new OpenClaw Architecture Deep Dive — Open-source agent architecture analysis
rewrite Ollama Local LLM Runner — Updated with Gemma 4, Qwen3-Coder, current model availability
rewrite RAG page → RAG Architecture Patterns 2026 — From basic vector indexing to full architecture guide
delete Hindsight Agent Memory (merged into Agent Memory Systems)
delete GitNexus Code Knowledge Graph (too shallow)
delete Superpowers Agent Framework (too shallow)
delete OpenClaw Personal AI Assistant (replaced by architecture deep dive)
delete MicroGPT, nanoGPT, llm.c, Autoresearch (merged into Karpathy Education Stack)
update Switched language convention to English
update All source links converted from wikilinks to plain text

ingest Raw/2026-03-18-ollama-local-llm-runner → created Ollama Local LLM Runner