How a Non‑English Seller Used a Bot and SearchCLI to Conquer Japan, Korea, and English‑Speaking Markets

Cross‑border sellers often stumble on language barriers: Japanese customers use polite long forms, Korean customers bargain with short phrases, and English customers demand precise material specifications. Hiring multilingual support staff is costly, and machine translation frequently fails.

Solution : The seller, called "V", used Kouzi to generate a web entry and Viking AI SearchCLI to create an intelligent multilingual bot that can understand Japanese, Korean, and English.

1. Quick Start – Installing SearchCLI

V sent the following command to Kouzi:

帮我下载这个 CLI：https://github.com/volcengine/SearchCLI ，并告诉我是否运行成功。

Within five minutes, Kouzi downloaded, installed, and authorized SearchCLI using an AK/SK obtained from the Volcano Engine console (first‑month trial costs ¥9.9).

2. Data Ingestion

SearchCLI splits data import into three steps:

Scan data and infer schema via vs item profile.

Generate an import plan with field mapping via vs item plan.

Upload data and build indexes using vs item apply.

This automated pipeline turned days of manual scripting into a single runnable flow, producing a searchable SKU base of over 2,000 items.

3. Multilingual Search, Recommendation, and QA

SearchCLI provides out‑of‑the‑box multimodal retrieval. In a Korean conversation, V queried “适合通勤的外套” and the bot returned relevant items by intent, not just keyword match.

For recommendation, V supplied product categories and a rough user persona; SearchCLI generated cold‑start simulated behavior data, enabling the recommendation engine to run without real interaction history.

In pre‑sale QA, the bot answered common questions about size, fabric care, shrinkage, and fit, eliminating the need for manual Q&A handling.

4. Evaluation and Automated Tuning

V used SearchCLI’s built‑in evaluation commands: vs search tune query-generate – automatically creates multilingual test queries from the indexed data. vs search tune run – runs batch scoring of different recall strategies. vs search tune report – produces a diagnostic report with actionable configuration suggestions. vs search tune apply – applies a new configuration and re‑evaluates.

After tuning, the recommendation repeat rate dropped 58% and novelty increased by 5.5%.

5. Building and Deploying the Storefront

V described the desired e‑commerce site to Kouzi (home, product list, detail, cart, checkout, and a floating bot). Kouzi instantly generated the pages, and the backend automatically connected to the Viking AI Search API with proper signing and SSE streaming.

Deploying required only clicking “Deploy” and adding four environment variables (the AK/SK). Kouzi provisioned an HTTPS public domain, handled DNS and certificates, and the site was live within seconds.

6. Summary

The seller achieved two key actions: (1) prepared the AI search, recommendation, and QA capabilities with SearchCLI; (2) built and deployed a multilingual storefront with Kouzi, all without writing a single line of algorithmic code.

Result: a fully functional, multilingual e‑commerce site that serves Japan, Korea, and English‑speaking markets, demonstrating that the right low‑code AI tools can replace language expertise and heavy development effort.