harshaAI / Prompt workflow application
NavigatorPrompt
A Next.js prompt workspace connected to FastAPI task APIs, Celery generation/evaluation workers, Redis result handling, and a saved prompt library.
View repository01 / COMPONENTS & DATA FLOW
Inside the system.
- 01Next.js action
- 02FastAPI task API
- 03Celery + Redis
- 04Status + prompt library
Task-based frontend contract
Server actions submit prompt-generation or evaluation work and receive a task ID and status URL. The frontend models PENDING, STARTED, SUCCESS, FAILURE, and RETRY states.
Background execution
The published Python API defines separate generation and evaluation queues, JSON task serialization, late acknowledgment, and retrying Celery tasks. Redis Sentinel configuration resolves master and replica clients.
History and library
The frontend database adapter uses PostgreSQL in production and a SQLite development fallback for local prompt history. Actions separate recent history from saved library entries and starring.
02 / IMPLEMENTATION DECISIONS
Why it works this way.
Long work has an identity
A task URL lets the interface inspect progress independently of the initial submission. The backend caches successful task results and exposes failure states.
Public code is a separate artifact
This repository contains a public prompt-workflow implementation. It is not presented as a published copy of UF IT’s internal NaviGator Prompt Builder deployment.
03 / SCOPE & TRADEOFFS
The implementation boundary.
The code demonstrates task APIs and persistence choices. Redis Sentinel configuration and retry settings alone do not prove a deployed high-availability topology or exactly-once execution.
04 / READ THE IMPLEMENTATION