Pinned Loading

1. lifegenie-claude-marketplace lifegenie-claude-marketplace Public

   LifeGenie personal plugin marketplace for Claude Code

   TypeScript

2. CulinaryAdvisor: A Two-System AI Arc... CulinaryAdvisor: A Two-System AI Architecture — Streaming chat (Gemini), multi-agent recipe pipeline (Claude SDK), character-driven personas, and batch image generation

   1

   # CulinaryAdvisor: A Two-System AI Architecture

   2

   3

   CulinaryAdvisor is a production AI platform for culinary mentorship and recipe generation. It runs as two independent systems that share a database and a persona layer.

   4

   5

   **Consumer App** (Next.js 16 + React 19 + Google Gemini)

3. Multi-agent research pipeline: 14 AI... Multi-agent research pipeline: 14 AI agents, 89 minutes, 1,991 lines of verified research on Gretel Beer (Austrian cuisine). Claude Code slash command, full output, and nano-banana image generation tool.

   1

   ---

   2

   name: research-chef

   3

   description: Research a chef from a taxonomy YAML using 4 sequential agent teams (2 researchers + 1 synthesizer each) to produce a completed research profile

   4

   argument-hint: "[path to taxonomy YAML, e.g., docs/taxonomy/european-cuisine-taxonomy.yaml]"

   5

   ---

4. Intake Is the Architecture: four int... Intake Is the Architecture: four intake pipelines (YouTube playlist, Discord URLs, manual upload, agent memory reconciliation) feeding a hybrid search layer (QMD) on commodity hardware — the recipe behind a Karpathy-LLM-Wiki-style personal KB. #personal-ai #claude-code #qmd #rag

   1

   # Intake Is the Architecture — Companion Recipe

   2

   3

   Technical deep-dive for the LinkedIn article *Intake Is the Architecture*. The article makes the argument. This gist gives the recipe.

   4

   5

   The thesis: Karpathy's [LLM Wiki](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f) names the destination (a markdown wiki maintained by an LLM). The system that makes the wiki actually useful is four intake pipelines, a hybrid search layer, and cross-machine serving — all running on commodity hardware. This gist walks through that system, piece by piece.

5. Claude Code Observability: OTEL to L... Claude Code Observability: OTEL to LangFuse — config, bridge setup, and start/stop commands. Full setup repo: github.com/lifegenieai/claude-code-observability #claude-code #langfuse #opentelemetry #observability #ai-agents

   1

   # Claude Code Observability: OTEL to LangFuse

   2

   3

   I pointed observability at my AI agents for the first time. Found 277,000 wasted tokens in thirty seconds of looking. This gist shows how the setup works. The full installable repo — with setup script, agent skill, and CLI patterns — is at **[lifegenieai/claude-code-observability](https://github.com/lifegenieai/claude-code-observability)**.

   4

   5

   Read the full story: [Turning on the lights](https://sapienttech.dev/posts/turning-on-the-lights/) on SapientTech.dev. Also available as a [video](https://youtu.be/cYmYhj6-Fpk).

6. The Council Protocol: Delphi-method ... The Council Protocol: Delphi-method framework for LLM-as-judge evaluation of creative AI output. Three judges, blinded Round 1, anonymized Delphi Round 2, pre-declared tiebreaks, full audit trail. Claude Code reference implementation with agent teams, domain-agnostic templates, and receipts from a real Opus 4.6 vs 4.7 production migration trial. #claude-code #llm-as-judge #delphi-method #ai-evaluation #agent-teams

   1

   # The Council Protocol: A Delphi-Method Framework for LLM-as-Judge Evaluation

   2

   3

   A reusable, auditable protocol for judging creative AI output on subjective dimensions (prose quality, voice consistency, cultural authority, narrative restraint, strategic fit) that no benchmark or schema check can measure.

   4

   5

   This Gist is the technical companion to the LinkedIn article *The Frikadeller Trial: How AI Judges Art*. The article tells the story of putting Claude Opus 4.7 on trial against 4.6 before migrating a production content pipeline. This Gist publishes the framework underneath, the Claude Code reference implementation, and the full audit trail from the trial itself.