Visual Roadmap — Part 2: Core Systems

LLM landscape, embeddings, vector search, RAG, agents, multi-agent systems, and MLOps.

5. The LLM Model Landscape (2026)

        flowchart TD
    subgraph Proprietary
        GPT[GPT-5.4 / o3 / o4-mini]
        CLAUDE[Claude Sonnet 4.6 / Opus 4.6]
        GEMINI[Gemini 3.1 Pro / Flash]
    end

    subgraph Open-Weight
        LLAMA[Llama 4 Scout / Maverick]
        QWEN[Qwen 3 0.6B-235B]
        DS[DeepSeek V3.2 / R1]
        PHI[Phi-4 14B]
        MISTRAL[Mistral / Mixtral]
    end

    subgraph Reasoning
        O3[OpenAI o3 / o4-mini]
        R1[DeepSeek R1 671B]
        CT[Claude Extended Thinking]
    end

    subgraph Local Running
        OLLAMA[Ollama]
        LLAMACPP[llama.cpp]
        MLX[Apple MLX]
        AITK[AI Toolkit for VS Code]
        OLLAMA --- PHI
        LLAMACPP --- QWEN
        MLX --- LLAMA
        AITK --- PHI
    end
    

6. Embeddings & Vector Search

        flowchart TD
    subgraph Embedding Models
        OAI_E["OpenAI text-embedding-3"]
        ST["Sentence Transformers"]
        COHERE_E["Cohere embed-v4"]
        NOM["nomic-embed-text"]
    end

    subgraph Vector Databases
        CHROMA[ChromaDB]
        QDRANT[Qdrant]
        PINECONE[Pinecone]
        WEAVIATE[Weaviate]
        PGVEC[pgvector]
        FAISS[FAISS]
    end

    DOC[Documents] --> CHUNK[Chunk] --> EMB[Embed]
    EMB --> OAI_E
    EMB --> ST
    OAI_E --> STORE[Store]
    ST --> STORE
    STORE --> CHROMA
    STORE --> QDRANT
    STORE --> PINECONE

    QUERY[Query] --> Q_EMB[Embed Query]
    Q_EMB --> SIM[Similarity Search]
    CHROMA --> SIM
    QDRANT --> SIM
    PINECONE --> SIM
    SIM --> RESULTS[Top-K Results]
    

7. RAG (Retrieval-Augmented Generation)

        flowchart TD
    subgraph Indexing Pipeline
        A[Documents] --> B[Chunk]
        B --> C[Embed]
        C --> D[Vector Store]
    end

    subgraph Query Pipeline
        E[User Question] --> F[Embed Query]
        F --> G[Retrieve Top-K]
        D --> G
        G --> H["Context + Question"]
    end

    subgraph Generation
        H --> I[LLM]
        I --> J[Answer with Citations]
    end

    subgraph "Advanced RAG — Query"
        L[Query Expansion] --> F
        HY[HyDE - Hypothetical Answers] --> F
    end

    subgraph "Advanced RAG — Retrieval"
        M[Hybrid Search BM25 + Dense] --> G
        K[Re-ranking / Cross-Encoder] --> G
        PC[Parent-Child Retrieval] --> G
        RAP[RAPTOR - Summary Trees] --> G
    end

    subgraph "Advanced RAG — Control"
        CR[CRAG - Retrieval Grading] --> H
        N[Agentic RAG] --> I
        GR[GraphRAG] --> G
    end
    

8. AI Agents Architecture

        flowchart TD
    USER[User Request] --> AGENT[Agent - LLM Brain]

    AGENT --> PLAN[Plan / Reason]
    PLAN --> ACT[Select Tool]
    ACT --> TOOL1["🔧 Web Search"]
    ACT --> TOOL2["🔧 Code Executor"]
    ACT --> TOOL3["🔧 Database Query"]
    ACT --> TOOL4["🔧 API Call"]

    TOOL1 --> OBS[Observation]
    TOOL2 --> OBS
    TOOL3 --> OBS
    TOOL4 --> OBS

    OBS --> REFLECT{Done?}
    REFLECT -->|No| PLAN
    REFLECT -->|Yes| ANSWER[Final Answer]

    subgraph Protocols
        MCP["MCP - Tool Connectivity"]
        A2A["A2A - Agent Delegation"]
    end

    subgraph Frameworks
        LC[LangChain / LangGraph]
        OAI_SDK[OpenAI Agents SDK]
        CREW[CrewAI]
        ADK[Google ADK]
        SK[Semantic Kernel]
    end
    

9. Multi-Agent Systems

        flowchart TD
    USER[User Task] --> COORD[Coordinator Agent]

    COORD --> R[Researcher]
    COORD --> W[Writer]
    COORD --> C[Critic]

    R -->|findings| COORD
    W -->|draft| COORD
    C -->|feedback| COORD

    COORD --> FINAL[Final Output]

    subgraph Patterns
        P1[Coordinator / Delegate]
        P2[Pipeline - Sequential]
        P3[Debate - Adversarial]
        P4[Voting - Consensus]
    end
    

10. MLOps Lifecycle

        flowchart TD
    A[Data Collection] --> B[Data Validation]
    B --> C[Feature Engineering]
    C --> D[Model Training]
    D --> E[Experiment Tracking - MLflow]
    E --> F[Model Evaluation]
    F --> G[Model Registry]
    G --> H[CI/CD Pipeline]
    H --> I[Deployment]
    I --> J[Monitoring & Observability]
    J --> K{Drift Detected?}
    K -->|Yes| A
    K -->|No| J

    subgraph Serving
        I --> S1[REST API - FastAPI]
        I --> S2[vLLM / TGI]
        I --> S3[Triton Inference Server]
        I --> S4[Managed APIs - Bedrock / Vertex AI / Azure AI Foundry]
        I --> S5[Edge Runtime - ONNX Runtime]
        I --> S6[Open Source - Ollama / llama.cpp / SGLang]
    end
    

← Previous: Overview | Next: Advanced Topics →