Build Your Own Translation Agent (No Code)

A reproducible pipeline with glossary and memory injection, eight automated checks that catch silent number errors, and a human review gate that never comes out.

A translation agent is not a chatbot you paste text into, it is a repeatable pipeline that applies your glossary and style rules, runs checks, and hands you a draft. This course builds one without code, and argues for a deterministic workflow over an autonomous agent because reproducibility is a professional requirement. You will learn the six-stage architecture, the injection techniques that separate a useful pipeline from a machine translation button (relevant glossary entries only, translation memory as few-shot examples), and eight automated checks including the number-set comparison that catches the silent figure changes LLMs make. It is direct about confidentiality (self-hosting the platform does not make the model local), tokenization cost on Indic scripts, and disclosure to clients.

What you'll learn

• Judge honestly when a translation pipeline is worth building at all
• Design the six-stage architecture without rebuilding a CAT tool
• Inject relevant glossary entries and TM matches for consistency
• Implement eight automated checks that fail loudly on real errors
• Handle confidentiality, tokenization cost and client disclosure properly
• Test against 20 real past jobs, and sell both the speed and the build

Course content

1. 1. What a translation agent is, and when it is worth building (15 min)
2. 2. The architecture, in plain terms (15 min)
3. 3. Injecting glossary, memory, and style (15 min)
4. 4. Automated checks that catch real errors (15 min)
5. 5. Confidentiality, cost, and disclosure (15 min)
6. 6. Building, testing, and selling it (15 min)

Related courses

• AI for Translators: Your First LLM Workflow — Drive ChatGPT, Claude & Gemini as a translation copilot: a repeatable, confidentiality-safe workflow that keeps you the editor in charge.
• Prompt Engineering for Translators — Turn vague AI output into on-brief translations: glossaries, style rules, and reusable templates that make ChatGPT, Claude and Gemini obey.
• Machine Translation Post-Editing (MTPE), Done Right — Post-edit to the right standard, light vs full per ISO 18587, spot the errors machines make, and price MTPE so it actually pays.