AI Product Builder (RAG) — AI Librarian for a Polish-language Knowledge Library

I run an established Polish-language personal development brand for professional women (YouTube, podcast, online programs). We are building an AI-powered knowledge library: hundreds of hours of video, audio, meditations and texts — all in Polish — organized and searchable, hosted on a community platform (Circle.so or similar), with two membership tiers. The heart of the product is an AI Librarian. A member types her real-life situation in her own words — “I can’t sleep”, “we keep fighting”, “I don’t know how to rest” — and the system responds like a good librarian would: “Start with this audio (12 min). Then this meditation. If you want to go deeper, this article.” 1–3 recommendations, with reasons and durations. The AI never invents teachings, never speaks as the founder, and is always transparent about being a tool. Its only job is to guide members through an existing, curated knowledge base. Scope (Version 1 — live September 1st): 1. Content pipeline: Polish audio/video transcription (e.g. Whisper) with QA, segmentation into logical units 2. AI-assisted tagging based on our existing taxonomy (4 content pillars + life situations — framework provided; we supply 20 hand-tagged reference materials for calibration) 3. Retrieval system (RAG) over the archive — semantic search that works well in Polish (this is a hard requirement; tell us how you’ll validate it) 4. Librarian interface: recommendation-style answers as described above, integrated into the community platform (widget/embed — no separate app) 5. Safety rules built in from day one: never first-person as the founder, always visibly a tool, crisis-related topics immediately redirected to a human and professional help, strict “not in the library” fallback instead of hallucination 6. Conversation logs reviewable by our team 7. Documentation + handoff: our assistant must be able to add new content monthly without you Version 1 should be simple and reliable — a smart recommendation system, not a fully conversational agent. Deeper conversationality is a possible Phase 2. Who we’re looking for An AI product builder who assembles systems from proven components — not an ML engineer (no model training needed) and not just a chatbot builder. You have shipped something like: an AI knowledge base, AI support agent over documents, internal AI assistant, or AI search. Required experience: RAG, LLM APIs (OpenAI / Anthropic / Gemini), vector databases (Qdrant, Pinecone, Weaviate, pgvector…), embeddings & semantic search, metadata/data architecture. Bonus: LlamaIndex / LangChain / n8n / Flowise, non-English retrieval, membership/community platforms, product thinking & UX. Process Milestone 1 is a paid pilot: 10 pieces of content through your full pipeline (transcription → tagging → retrieval → sample librarian answers to 15 test questions in Polish — we provide the questions and evaluate with native speakers). If quality is right, we proceed to the full archive and integration. NDA required before receiving any content — the archive is the core asset of this business. At handoff we own everything: accounts, code/workflows, documentation. In your proposal, please answer: 1. Links/descriptions of 1–2 similar systems you’ve built (archive → knowledge base with retrieval) 2. Your recommended stack for this project — and why 3. How will you ensure the librarian only uses our materials and says “I don’t have this in the library” instead of inventing answers? 4. How will you make retrieval work well in Polish, and how will you test it? 5. What do you think is the biggest challenge in this project? 6. Fixed price broken into milestones (pilot priced separately) + realistic timeline Template answers will be ignored. We care about the right person, not the cheapest offer — but we work milestone by milestone, not on open-ended retainers.