Summary

• wire the backend dataset/search repository to prebuilt RAGU indexes and MixSearchEngine
• update API schema artifacts and chat UI contract for the RAGU search response shape
• document every supported backend environment variable in back/.env.example, including YandexGPT, OpenAI-compatible LLM fallback, and embedding endpoint options
• stabilize the MAM CI TypeScript version used by deploy workflow

Why

The backend now depends on existing RAGU index artifacts instead of mock-only retrieval paths. The environment example had fallen behind the supported configuration surface, so new fallback LLM and embedder variables were missing from onboarding docs.

Validation

• comm -23 <(rg -h -o --pcre2 'env\\.get\\(\"\\K[A-Z0-9_]+' back/src | sort -u) <(rg -h -o --pcre2 '^[A-Z0-9_]+(?==)' back/.env.example | sort -u) produced no missing variables
• git diff --check -- back/.env.example
• RAGU_INDEXES_DIR=/home/yara/Desktop/dev/other/RAGU/NorHac/RAGU/indexes .venv/bin/python -m pytest from back/: 9 passed, 1 warning

Notes

A plain .venv/bin/python -m pytest run without a local RAGU_INDEXES_DIR currently fails because dataset discovery returns an empty list in this checkout.