NEXORA — AI Knowledge Workspace

Full-stack SaaS platform that turns uploaded documents into a searchable knowledge base — ask questions in plain English and get grounded RAG answers with source citations.

Problem Statement

Teams drowning in PDFs, contracts, and reports need a way to query their documents without manually searching through files. Generic chatbots hallucinate without source grounding.

Solution

Built NEXORA — a multi-tenant AI knowledge workspace with document ingestion, vector search, streaming RAG chat, team collaboration, usage analytics, and tiered billing. Runs in demo mode out of the box or connects to MongoDB, Redis, S3, and OpenAI for production.

Features

Intelligent document ingestion — auto-extract, chunk, and embed PDF, DOCX, TXT, Markdown, and CSV files
Streaming RAG chat with source citations and conversation history
Semantic search across all workspace documents
Multi-tenant workspaces with role-based access (Owner, Admin, Member, Viewer)
Background job pipeline via BullMQ — extract, index, and notify on completion
Analytics dashboard, team invitations, billing tiers (Free, Pro, Team), and dark/light theme

My Contributions

Architected the full-stack application — Next.js App Router, service layer, and MongoDB models
Built the RAG pipeline — text extraction, chunking, vector search, and AI provider abstraction
Implemented BullMQ workers for async document processing and notifications
Designed the landing page, authenticated app shell, and billing/analytics dashboards

Challenges

Designing a document processing pipeline with status tracking from upload through embedding
Building streaming RAG chat with NDJSON responses and real-time source references
Implementing workspace-scoped RBAC and rate limiting across server actions and API routes