I need a web-based application—“CPC’s Friendly Water Expert”—that lets me upload our PDFs, Excel files, and other product documents, then answers user questions strictly from that material. My customers will be the main users, though I also want internal and partner logins, so role-based access (admin, internal, partner) is essential; the admin role only needs moderate control such as approving uploads, managing user accounts, and viewing usage analytics. Key flow • An upload area where an admin can drag-and-drop new PDFs or spreadsheets. As soon as a file lands, it should be parsed, embedded, and added to the knowledge base. • A chat interface that accepts natural-language questions and returns clear answers plus inline citations (“Pulled from Water-Softener-Guide.pdf p. 7”). If the answer spans several sources, list each one. • A document library so anyone with permission can browse or download original files. Quick filters by product line, format, or upload date will help. • Branding that matches CPC’s water-treatment look—clean blues, plenty of white space, and our logo in the header and chat avatar. Technical expectations – Real-time retrieval-augmented generation (RAG) so responses never drift beyond my data set. – Secure storage of embeddings (for instance, PostgreSQL + pgvector, Pinecone, or equivalent). – Modern stack—React or Vue for the front end, Node, Python FastAPI, or similar on the back end—open to your preference as long as it is maintainable. – Authentication using JWT or OAuth and an RBAC layer that enforces the three roles. – Deployment ready for a standard cloud environment such as AWS, Azure, or GCP. Deliverables 1. Source-controlled codebase with README for local setup. 2. Fully styled upload page, chat page, and document library. 3. Admin dashboard with the moderate controls described. 4. Initial onboarding of a small sample data set to prove end-to-end accuracy. 5. Short hand-off call or video explaining architecture and future scaling considerations. Acceptance criteria • Chat answers reference only uploaded material and always show at least one citation. • Non-admins cannot see upload controls. • Library downloads respect role permissions. • UI follows CPC brand guide (to be supplied). • Tests demonstrate file ingestion, RAG accuracy, and permission checks. When you respond, focus on your relevant experience building RAG chatbots, secure role-based apps, or similar document Q&A systems. Links or screenshots of live projects are perfect.