Applied AI / Nuclear document workflow

NuclearFlowAI

An in-progress applied AI workflow demo for nuclear documents, kept in a private repository for now, using public CNSC reactor licensing guidance as the example document set.

Status Private repository while in development

Stack Python, FastAPI, React, Next.js, SQLite, RAG patterns

Focus Cited retrieval, deterministic responses, saved findings, exports, audit logs

Scope Public upload, OCR, ingestion, index rebuilds, and admin controls are disabled

Project notes

This is still being built, so the public-facing description stays limited to the architecture and workflow decisions that are stable enough to describe now.

The design starts from traceable nuclear work, not chat.
The demo is designed around cited source excerpts, saved findings, and summary export rather than a free-form chatbot experience.
The current demo uses seeded public CNSC guidance.
Visitors can browse prepared public documents, run search, ask grounded questions, save useful findings, and export a sample summary.
The workflow is being used to test RAG, GraphRAG, and OmniRAG patterns.
The local demo supports deterministic cited responses, retrieval traces, saved finding decisions, audit events, and capped public usage so answer quality can be inspected rather than guessed.
The backend work is deliberately product-shaped.
The current scope includes API routes, workflow state, saved findings, export behavior, validation, and limits around what public users can run.
The public workflow is deliberately restricted.
Public users cannot upload files, trigger OCR, run ingestion, rebuild indexes, create projects, or access admin controls.
The next work is deployment-oriented infrastructure.
Planned work includes Postgres and Redis-backed persistence, Weaviate indexing, and private admin ingestion behind admin-only controls.