Reducing QC Story Documentation Effort by 80%+ Through AI-Guided Automation

Industry Overview

Quality improvement programs in manufacturing depend on structured problem-solving and consistent documentation. However, when Quality Circle knowledge remains buried in individual documents, teams repeatedly spend expert time recreating similar analyses instead of building on previous investigations.

The Problem

At TVS Motor Company, creating each QC Story required significant expert effort, with structured content repeatedly drafted from scratch.

Approved QC Stories and their root causes were stored in PPT files, making historical knowledge difficult to search and reuse during new investigations.

The process also lacked a consistent compliance gate across teams. Stories could vary in methodology adherence, root-cause specificity, and target definition. There was also no complete audit trail showing what had been drafted, reviewed, rejected, or revised.

Our Role

Ganit was tasked with building an AI-guided platform that could automate QC Story drafting, enforce Quality Circle methodology compliance, and surface relevant historical knowledge while keeping human authors in control of critical decisions.

Methodology

Ganit developed a conversational, agent-orchestrated application on Databricks that guides QC authors through each stage of the QC Story process.

The platform uses five specialised AI agents:

• Drafter Agent — Generates structured QC Story sections
• Scoring Agent — Evaluates compliance with QC methodology
• RAG Agent — Surfaces relevant historical QC Stories
• Pareto Agent — Prioritises factors through Pareto analysis
• Diagnostic Agent — Validates root causes with statistical evidence

A 20+ node LangGraph state machine orchestrates the workflow. Each section is drafted by AI but requires explicit human acceptance before the process can advance.

Root causes are validated using statistical techniques such as Cohen's d and Cramér's V, producing Strong, Medium, or Weak evidence-based verdicts rather than relying on LLM judgement alone.

Historical QC Story presentations are indexed using Databricks Vector Search so relevant past investigations can automatically surface during problem identification.

How It Works

Author Prompt → Agent Drafting → RAG Retrieval → Cause Scoring → Human Review → Delta Logging

The architecture spans problem identification, observation, root-cause analysis, and final review, with human checkpoints and specialised agents embedded throughout the process.

Our Impact

The solution reduced manual QC Story documentation effort by approximately 80%+.

Root causes are now backed by statistical evidence, while every draft, rejection, revision, and LLM interaction is logged to Delta tables in real time. A scoring gate prevents the process from advancing until compliance requirements are met.

Approved QC Stories also become part of the knowledge base, allowing organisational knowledge to accumulate and support future investigations instead of remaining locked inside individual presentations.