Skip to content

Nsujatno/hackutd25

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

26 Commits
 
 
 
 
 
 

Repository files navigation

Kanban Sync: AI-Powered Workflow & Validation for Datacenters

HackUTD Hackathon Submission | November 2025

Project Overview

Kanban Sync is a sophisticated, AI-powered workflow assistant engineered to dramatically improve the efficiency and safety of datacenter operations. In complex environments where minor errors can lead to major outages, the project's core mission is twofold:

  1. Natural Language Task Creation: Allow engineers to create, describe, and manage complex maintenance tasks using natural language, translating high-level intent into actionable steps.

  2. Pre-Execution Validation: Implement a safety-critical AI pipeline to validate all natural language instructions against official datacenter operational manuals, preventing non-compliant or potentially dangerous actions before they are executed.

Kanban Sync transforms manual, error-prone checklist execution into a streamlined, context-aware, and safety-validated workflow.

Workflow (dual RAG pipeline)

The central innovation of Kanban Sync is its Dual-RAG (Retrieval-Augmented Generation) pipeline, designed for rigorous instruction validation:

  1. The Validation Challenge

Datacenter environments are governed by extensive, constantly updated documentation. The challenge is ensuring that a newly created task—described in conversational language by an operator—does not conflict with established procedures, security protocols, or hardware limitations buried deep within manuals.

  1. The Dual-RAG Solution

We implemented a two-stage RAG system leveraging OpenAI models for precision and efficiency:

Embedding/Retrieval: We utilized OpenAI Ada embeddings to convert the comprehensive set of datacenter manuals into high-density vector embeddings stored in a Supabase vector database. This allows for rapid and contextually accurate similarity searching.

Validation & Generation: When a task is submitted, the system retrieves the most semantically relevant sections of the manuals. This context is then fed, along with the user's task instruction, to GPT-4o-mini. The model's task is to analyze the retrieved manual sections and explicitly confirm or deny the validity of the proposed instruction.

This dual-system approach ensures validation is fast, highly accurate, and directly grounded in the source documentation, effectively serving as an intelligent safety gate.

Tech stack

FastAPI

  • High-performance API framework handling all input/output. It serves as the gateway for processing natural language tasks and orchestrating the RAG pipeline calls.

Supabase

  • Used for complete backend infrastructure, relational data storage (Kanban board, user profiles), and hosting the vector database necessary for the RAG pipeline's similarity search.

GPT 4o mini

  • The core LLM responsible for analyzing user instructions, querying the retrieved documents, and providing the final, validated decision (compliant/non-compliant) for the task.

OpenAI ada embeddings

  • The standard embedding model used to index the datacenter manuals into vector format, powering the highly efficient contextual retrieval.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 3

  •  
  •  
  •