# `PhoenixKitProjects.AITranslatable`
[🔗](https://github.com/BeamLabEU/phoenix_kit_projects/blob/v0.14.0/lib/phoenix_kit_projects/ai_translatable.ex#L1)

`PhoenixKitAI.Translatable` adapter for projects resources —
the small per-module hook into PhoenixKitAI's generic AI-translation pipeline.

Serves four resource types — `"project"`, `"template"` (a `Project` with
`is_template: true`), `"task"`, and `"assignment"` — each translating the
fields its schema declares via `translatable_fields/0` (project/template:
`name` + `description`; task: `title` + `description`; assignment:
`description`).

## Storage

Translations live in each schema's `translations` JSONB column, shaped
`%{lang => %{field => value}}` with **plain** field keys (no underscore
prefix — unlike catalogue's multilang `data`). Source text comes from
`translations[source_lang][field]` when present, else the resource's
primary column (`name`/`title`/`description`). Because the per-language
map stores values verbatim, an AI result identical to the source (a code,
text already in the target language) is still persisted — the field fills
in and the language stops reading as "missing", with no extra handling.

Registered via the `PhoenixKitProjects.ai_translatables/0` function, which
PhoenixKitAI discovers by duck-typing. The enqueue, the
AI call, the per-resource PubSub broadcasts, the retry policy, and the
audit log all live in PhoenixKitAI — this module only reads source fields
and merges results back atomically.

---

*Consult [api-reference.md](api-reference.md) for complete listing*