[MISSION] RAG system for an endangered spoken language — 10 weeks — full IP transfer

Hi, I’m looking for an experienced NLP/LLM engineer for a serious, well-scoped project: building the first RAG-based localization engine for a low-resource language spoken in South America.

The project is built on a proprietary corpus (pedagogical content, dictionary, proverbs, orthographic rules) developed over 4 years. Architecture, endpoints, quality criteria and deliverables are fully specified in a detailed technical brief (available under NDA).

**Technical scope (MVP — 10 weeks)**

- RAG pipeline on low-resource language corpus (LangChain or LlamaIndex)
- Multilingual embedding model (multilingual-e5 or equivalent)

• Vector DB: Pinecone, Weaviate or Supabase pgvector — latency < 500ms
• Modular prompt layer — 6 use-case templates (translate, educate, dub, subtitle, localize, campaign)
• Multi-tenant B2B SaaS infrastructure — strict data isolation, JWT auth, configurable quotas
• REST API + Swagger documentation
• Admin interface for glossary management (no-dev updates)
• Offline SQLite bundle for React Native mobile app

Profile required

• Proven RAG experience on low-resource or multilingual corpora
• Multi-tenant SaaS architecture references
• Modular prompt engineering in production (Anthropic or OpenAI API)
• Full IP transfer to client — non-negotiable
• NDA before corpus access
• Budget: 5,000–10,000€ depending on experience

- Start: within 2–3 weeks

If this matches your profile, please reply with concrete references on RAG low-resource languages, multi-tenant SaaS architecture, and modular prompt engineering in production