{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "716e0803-982b-49eb-926f-4d7bc1d24c71",
   "metadata": {},
   "source": [
    "# 5 Workflow Design Pattern"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "13d96dc6",
   "metadata": {},
   "source": [
    "### 1. Prompt Chaining : \n",
    "Result from one llm to another llm"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "633cd510-5697-4890-a527-c36942a1c2d0",
   "metadata": {},
   "source": [
    "```mermaid\n",
    "flowchart LR\n",
    "    IN((IN)):::in --> LLM1[LLM1]:::llm\n",
    "    LLM1 --> Gate[Gate]:::gate\n",
    "    Gate --> LLM2[LLM2]:::llm\n",
    "    LLM2 --> LLM3[LLM3]:::llm\n",
    "    LLM3 --> OUT((OUT)):::out\n",
    "\n",
    "    classDef in fill:#ff7f0e,stroke:#000,color:#fff;\n",
    "    classDef out fill:#ff7f0e,stroke:#000,color:#fff;\n",
    "    classDef llm fill:#b58900,stroke:#000,color:#fff;\n",
    "    classDef gate fill:#1f77b4,stroke:#000,color:#fff;\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bf9dc153-87a4-4a1a-b045-488408b2b944",
   "metadata": {},
   "source": [
    "### 2. Routing\n",
    "One LLM acts as a router to route to specialist llms"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1c5e9561-643f-4368-8c35-94482134ddca",
   "metadata": {},
   "source": [
    "```mermaid\n",
    "flowchart LR\n",
    "    IN((IN)):::in --> Router[LLM Router]:::gate\n",
    "    Router --> LLM1[LLM1]:::llm\n",
    "    Router --> LLM2[LLM2]:::llm\n",
    "    Router --> LLM3[LLM3]:::llm\n",
    "    LLM1 --> OUT((OUT)):::out\n",
    "    LLM2 --> OUT\n",
    "    LLM3 --> OUT\n",
    "\n",
    "    classDef in fill:#ff7f0e,stroke:#000,color:#fff;\n",
    "    classDef out fill:#ff7f0e,stroke:#000,color:#fff;\n",
    "    classDef llm fill:#b58900,stroke:#000,color:#fff;\n",
    "    classDef gate fill:#1f77b4,stroke:#000,color:#fff;\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "56c725b5-38b0-4cc0-8713-5a5d300e5a5c",
   "metadata": {},
   "source": [
    "### 3. Parallelization\n",
    "Code as a router and in parallel sent to llms for answer and then again code aggregates and send final output "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "555129f1-3d4f-46b3-a154-afb8331508ef",
   "metadata": {},
   "source": [
    "```mermaid\n",
    "flowchart LR\n",
    "    IN((IN)):::in --> Coord[Coordinator]:::gate\n",
    "    Coord --> LLM1[LLM1]:::llm\n",
    "    Coord --> LLM2[LLM2]:::llm\n",
    "    Coord --> LLM3[LLM3]:::llm\n",
    "    LLM1 --> Agg[Aggregator]:::gate\n",
    "    LLM2 --> Agg\n",
    "    LLM3 --> Agg\n",
    "    Agg --> OUT((OUT)):::out\n",
    "\n",
    "    classDef in fill:#ff7f0e,stroke:#000,color:#fff;\n",
    "    classDef out fill:#ff7f0e,stroke:#000,color:#fff;\n",
    "    classDef llm fill:#b58900,stroke:#000,color:#fff;\n",
    "    classDef gate fill:#1f77b4,stroke:#000,color:#fff;\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "65b5521e-1f5f-4e96-90f6-d85b8f2b4624",
   "metadata": {},
   "source": [
    "### 4. Orchestrator-Worker\n",
    "Similar to Parallelization but instead of code, it is LLM"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ed59e3a1-a64f-4619-b051-5fa42575f56a",
   "metadata": {},
   "source": [
    "```mermaid\n",
    "flowchart LR\n",
    "    IN((IN)):::in --> Orchestrator[Orchestrator]:::llm\n",
    "    Orchestrator --> LLM1[LLM1]:::llm\n",
    "    Orchestrator --> LLM2[LLM2]:::llm\n",
    "    Orchestrator --> LLM3[LLM3]:::llm\n",
    "    LLM1 --> Synth[Synthesizer]:::gate\n",
    "    LLM2 --> Synth\n",
    "    LLM3 --> Synth\n",
    "    Synth --> OUT((OUT)):::out\n",
    "\n",
    "    classDef in fill:#ff7f0e,stroke:#000,color:#fff;\n",
    "    classDef out fill:#ff7f0e,stroke:#000,color:#fff;\n",
    "    classDef llm fill:#b58900,stroke:#000,color:#fff;\n",
    "    classDef gate fill:#1f77b4,stroke:#000,color:#fff;\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "57c7109b",
   "metadata": {},
   "source": [
    "### 5. Evaluator-Optimizer\n",
    "LLM Output is evaluated by another LLM and it either accepts or reject the answer"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "439b4e7d-1009-418a-b88c-a4f5bd089bd5",
   "metadata": {},
   "source": [
    "```mermaid\n",
    "flowchart LR\n",
    "    IN((IN)):::in --> Generator[LLM Generator]:::llm\n",
    "    Generator --> Evaluator[LLM Evaluator]:::llm\n",
    "    Evaluator -->|Accepted| OUT((OUT)):::out\n",
    "    Evaluator -.->|Rejected with Feedback| Generator\n",
    "\n",
    "    classDef in fill:#ff7f0e,stroke:#000,color:#fff;\n",
    "    classDef out fill:#ff7f0e,stroke:#000,color:#fff;\n",
    "    classDef llm fill:#b58900,stroke:#000,color:#fff;\n",
    "    classDef gate fill:#1f77b4,stroke:#000,color:#fff;\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "97096f3a-0c5f-4ff8-9b20-7cb7fce2195f",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.12.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}