I built an offline Mongolian drafting tool for border-town merchants. Here’s where it breaks.

Description and use

CLI merchant‑ai‑draft generates Cyrillic Mongolian drafts from Chinese merchant JSON, but struggles with terminology and currency.

Problem

Professional translators are slow, expensive, and sometimes unreliable for merchants in Erenhot who need written product information in Mongolian Cyrillic.

Outcome

The CLI produces native Cyrillic Mongolian drafts in about 3 seconds per sentence on Apple Silicon, but struggles with terminology, currency units, and product clauses. Photo‑to‑description works well in Chinese but translation to Mongolian mangles key terms.

Workflow

• 1. Prepare a merchant JSON file in Chinese.
• 2. Run python mn_draft.py --input merchant.json to generate a draft in Cyrillic Mongolian.
• 3. Review the draft; if marked reviewed, the file remains unchanged. The tool enforces AI proposal and human decision in code.
• 4. Use the output as the basis for a bilingual electronic product card.
• 5. Optionally, spike a photo‑to‑description flow with qwen2.5vl:7b for product images.

Evidence

merchant‑ai‑draft — an offline CLI that takes a merchant JSON file (Chinese copy), fills in Mongolian (Cyrillic) first drafts with a local open‑weight model, and marks everything draft.

Source

NLLB-200-3.3B (selected) outputs Cyrillic Mongolian natively — but terminology, currency units, and clauses remain unreliable. ~3 seconds per sentence on Apple Silicon.

Source

The CLI itself is covered by 11 automated tests (review protection, ID preservation, offline failure handling).

Source