I would like you to go to danielkliewer.com and github.com/…

I would like you to go to danielkliewer.com and github.com/kliewerdaniel and your purpose is to return the repos which would be good to have cloned in a folder to help plan out the following by giving it more context for patterns which I have previously done : ### **Project Overview: The Simulacra System** The project aims to create a **self-organizing idea lab** where every concept or insight is stored as a structured node in a schema-driven knowledge graph. Unlike static chatbots, this system is designed to **evolve over time**, reflecting a "round character" that adapts its worldview based on external data (RSS feeds) and user interactions. Its primary emotional and technical capstone is the **"Chris Bot,"** a digital resurrection of a deceased friend through the synthesis of years of personal data and specialized persona engineering. --- ### **Technical Specification (The "Superstack")** The author mandates a **local-first philosophy**, prioritizing privacy and data sovereignty by running all compute on personal hardware to avoid cloud costs and surveillance. | Layer | Technology Choice | Rationale | | :--- | :--- | :--- | | **Frontend** | **Next.js 14/16 (App Router)** | Provides a modern, component-driven UI for real-time visualization and interaction. | | **Backend** | **FastAPI or Django** | Handles high-performance asynchronous orchestration and manages structured database endpoints. | | **LLM Inference** | **Ollama or llama.cpp** | Serves local models (e.g., Mistral, Qwen2.5, Llama 3) for private, zero-cost processing. | | **Graph Database** | **Neo4j or NetworkX** | Stores relational entities and "memories" as a mind map of interconnected concepts. | | **Vector Database** | **ChromaDB or FAISS** | Facilitates semantic search and Retrieval-Augmented Generation (RAG). | | **Multimodal** | **ComfyUI, Stable Diffusion, Coqui TTS** | Generates contextual images and voice synthesis for an immersive narrative experience. | --- ### **Core Features and Modules** #### **1. Quantified Persona Engine** The system extracts a **numerical style fingerprint** from writing samples. * **Persona Schema:** A JSON/YAML dictionary consisting of **50 distinct psychological and stylistic traits** (e.g., skepticism, empathy, vocabulary complexity) rated on a 0.0 to 1.0 scale. * **Dynamic Prompting:** These weights populate f-strings in system prompts, ensuring the LLM adopts a consistent, specific voice. * **Trait Evolution:** Traits are adjusted dynamically via **Reinforcement Learning from Human Feedback (RLHF)** using sliders in a "character creation" style UI. #### **2. GraphRAG Knowledge Ingestion** The system transforms unstructured data (.md files, PDFs, Reddit scrapes) into a **semantic knowledge substrate**. * **Smart Ingestion:** Uses a **sliding window chunking strategy** to maintain context while extracting entities and relationships as graph nodes and edges. * **Hybrid Search:** Combines dense vector similarity search with graph traversal (using algorithms like PageRank) to find contextually relevant "memories". * **Provenance:** Every generated response must cite its source within the graph to minimize hallucinations. #### **3. Multi-Agent Orchestration** Instead of a single prompt, the system uses a **Standard Operating Procedure (SOP)** to coordinate specialized agents arranged in a Directed Acyclic Graph (DAG). * **Agent Roles:** Includes a **Researcher** (retrieval), **Writer** (synthesis), **Critic** (evaluation), and **Orchestrator** (workflow management). * **Continuous Feedback Loop:** Each agent's output is validated against established guidelines (e.g., `ai_guidelines.md`) before moving to the next node. #### **4. Multimodal Synthesis & "Live" Feeding** The system creates a **digital heartbeat** through constant input/output loops. * **Infinite Newsfeed:** Scrapes RSS feeds, clusters them by topic using K-means, and uses the active persona to generate satirical or objective news segments. * **Visual Storytelling:** Uses LLaVA to analyze images or current events and Stable Diffusion to generate matching artistic representations. * **Voice Embodiment:** Clones the user's or persona's voice (e.g., "Chris") for a real-time, narrated "AI radio" experience. --- ### **Implementation Roadmap** 1. **Environment Setup:** Dockerize the stack including Neo4j, ChromaDB, and an Ollama container with GPU support. 2. **Documentation Anchoring:** Create a `/docs` folder containing the "North Star" specs (`architecture.md`, `persona_schema.md`, `api_spec.md`). 3. **Core Pipeline:** Build the Python ingestion module that parses .md files into the graph and vector databases. 4. **Persona Integration:** Implement the two-step LLM call (metadata extraction → persona generation) to populate the trait vector. 5. **Frontend Dashboard:** Develop the Next.js interface featuring **ReactFlow** for graph visualization and **shadcn Sliders** for trait manipulation. 6. **RLHF Layer:** Integrate a logging mechanism (`ai_output.md`) that acts as a ledger to track agent decisions and allow for user-driven refinement.