API Reading

Keypoint

• Authentication(How to use the header?)
• Content type(JSON or Multipart)
• Required or Optional parameter

status code

429(too many request)
500

Q1 RemoveBackground

題目描述： 請閱讀以下 API 文件摘要，寫一段 Pseudocode 函式 RemoveBackground(image_data)，回傳去背後的圖片網址。若失敗請回傳 Error。

API Endpoint: POST https://api.vivipic.com/v1/remove-bg

Headers:

• Authorization: Bearer
• Content-Type: multipart/form-data Body

Parameters:

• file: The image file (Binary). (Required)
• output_format: “png” or “webp”. Default is “png”. (Optional)

Response (JSON):

• Success (200): { “status”: “success”, “url”: “https://…” }
• Error (4xx/5xx): { “status”: “error”, “message”: “Invalid file” }

A1

code
FUNCTION RemoveBackGround(image_binary)
    // 1. Setup configuration
    api_url = https://api.vivipic.com/v1/remove-bg
    api_key = GET_ENV("VIVIPIC_API_KEY")

    // 2. prepare header
    header = {
        "Authorization": "Bear" + api_key,
        "Content-Type": "multipart/form-data Body"
    }

    // 3. prepare payload
    payload = {
        "file": image_binary,
        "output_format": "png" //explicitly setting optional parameter
    }

    // 4. request
    response = HTTP(api_url, header, payload)

    // 5. handling status code
    status_code = response.status_code
    IF status_code == 200 THEN
        json_body = PARSE_JSON(response.body)
        RETURN json_body.url
    ELSE IF status_code == 402 THEN
        RETURN ERROR("Unauthorized: check API Key")
    ELSE IF status_code == 500 THEN
        RETURN ERROR("Server Error: please try again later")
    ELSE
        //capture generic error
        json_body = PARSE_JSON(response.body)
        RETURN ERROR("API Failed" + json_body.message)
    END IF
END FUNCTION

Q2 AI 服務串接與錯誤處理 (AI Pipeline Orchestration)

對應 JD： 定義APP和Server的溝通協定、導入實際手機APP程式、去背景

核心考點： API 串接邏輯、Promise/Async Chain、錯誤復原 (Fallback)。

情境： App 要執行一個「人像風格化」的功能，流程如下：

• 呼叫 RemoveBG_API 去背。
• (成功後) 將去背圖傳給 StyleTransfer_API 進行風格化。
• (成功後) 將結果回傳。 挑戰： 如果第 1 步去背失敗，不要報錯，而是直接拿原圖去跑第 2 步（Fallback 機制）。請設計這個流程。

A2

/*
Function: ProcessPortraitStyle
Description: Orchestrates a multi-stage AI pipeline with fallback logic
*/

FUNCTION ProcessPortrainStyle(ori_image) -> Image
    current_image = ori_image

    // 1. try remove background
    TRY
        // Assume CALL_API is async function that throws error on 4xx/5xx
        api_url = "https://api.vivipic.com/remove-bg"
        bg_remove_img = AWAIT CALL_API(api_url, ori_image)

        // If successful
        current_image = bg_remove_img
    CATCH Error e
        // Fallback logic
        LOG("Remove BG error" + e.message)
        current_image = ori_image
    END TRY

    // 2. style transfer
    TRY
        api_url_transfer = "https://api.vivipic.com/style-transfer"
        final_result = AWAIT CALL_API(api_url_transfer, current_image)

        RETURN currrent_image
    CATCH Error e
        THROW Error("Transfer failed: " + e.message)
    END TRY
END FUNCTION

/*
RESPONSE JSON SCHEMA DESIGN:
{
    "task_id": "12345",
    "status": "completed", // or "partial success", "failed"
    "data": {
        "image_url": "https://...",
        "processing_details": {
            "remove_bg": "skipped_error", //Client knows BG removal failed
            "style_transfer": "success"
        }
    }
}
*/

「在設計 AI Pipeline 時，容錯性 (Fault Tolerance) 非常重要。針對去背功能，我設計了 Fallback 機制：如果去背 API 掛了或超時，系統會自動『降級服務 (Graceful Degradation)』使用原圖繼續後續流程，而不是直接讓 User 看到錯誤畫面。同時，我在 API 回傳的 JSON 結構中定義了 processing_details，讓 App 端知道發生了什麼事。」