ASR Interview Coder LLM System Flow

System Design · flowchart diagram · NOASSERTION

This diagram illustrates the four-layered processing flow for an ASR-driven LLM system, including smart analysis, intent recognition, user personalization,

Source: https://github.com/lanzeweie/ASR-interview-coder-llm/blob/cad6297ac1dc16063215e57c175239b2019d44ab/README/Mermaid.md
Curated by lanzeweie
ASR LLM System Design Conversational AI Agents Flowchart AI

Mermaid source

flowchart TD
    %% --- 系统入口 ---
    subgraph Entrances [系统入口]
        ASR([ASR消息输入])
        Manual([手动输入消息])
    end

    %% 第一层入口：ASR自动触发 或 手动消息
    ASR --> SmartCheck{智能分析开启？}
    Manual --> L1_Manual[手动进入第一层]

    %% 第一层：智能分析（可关闭）
    SmartCheck -- 否 --> L1_Manual
    SmartCheck -- 是 --> SmartTrigger[第一层：智能分析 Agent]
    SmartTrigger --> SmartResult{智能分析返回？}

    SmartResult -- false --> ContinueListen[继续监听]
    SmartResult -- true --> L1_Out[第一层完成]

    L1_Manual --> IntentCheck
    L1_Out --> IntentCheck

    %% 第二层：意图识别（可关闭）
    IntentCheck{意图识别开启？} -->|否| L3_Start
    IntentCheck -->|是| IntentAgent[第二层：意图识别 Agent]
    IntentAgent --> L2_Out[意图结果]
    L2_Out --> L3_Start

    %% 第三层：用户个性化（可关闭）
    L3_Start --> PersonalCheck{用户个性化开启？}
    PersonalCheck -- 否 --> L4_Start
    PersonalCheck -- 是 --> Personal[第三层：用户个性化]
    Personal --> L4_Start

    %% 第四层：回答生成（必选）
    subgraph Answering [第四层：回答生成]
        L4_Start --> ModeCheck{智囊团模式？}

        ModeCheck -- 是 --> ThinkTank[智囊团并行模型回答]
        ModeCheck -- 否 --> SingleModel[单模型回答]

        ThinkTank --> Collect[加载目标岗位提示词]
        SingleModel --> Collect
        Collect --> ENDs[输出]
    end

What this diagram shows

This flowchart details a four-layered system architecture for processing ASR or manual inputs using LLMs and agents. It begins with message input (ASR or manual), proceeds through an optional smart analysis layer, followed by optional intent recognition and user personalization layers. The final layer focuses on answer generation, offering a choice between a "Think Tank" (parallel model) approach or a single model, before loading prompts and producing an output.

When to use it

Ideal for designing conversational AI systems, voice assistants, or automated interview platforms that require multi-stage processing, context awareness, and flexible answer generation strategies. Useful for systems integrating ASR with LLM agents.

How to adapt it for your project

This architecture can be adapted by adding more layers for complex pre-processing or post-processing, integrating different types of LLM agents for each layer, or customizing the "Think Tank" with specific domain models. The toggles for smart analysis, intent recognition, and personalization allow for flexible deployment based on specific use cases.

Key concepts