Service Work Order AI Pipeline

Service Work Order AI Pipeline A data-flow diagram generated by Archify. 01 / Source 02 / Extract 03 / Curate 04 / Analyze 05 / Act PostgreSQL · work orders · 01 / Source · 6-7k tickets PostgreSQL work orders 6-7k tickets Batch Export · Fabric / SQL · 01 / Source · read-only Batch Export Fabric / SQL read-only LLM Extractor · 3-4k tok/ticket · 02 / Extract · structured fields LLM Extractor 3-4k tok/ticket structured fields Human Review · sampled QA · 03 / Curate · convergence check Human Review sampled QA convergence check Structured Store · curated records · 03 / Curate · labeled Structured Store curated records labeled Material Master · normalized PNs · 03 / Curate · FPGrowth Material Master normalized PNs FPGrowth Power BI · DAX reports · 04 / Analyze · bathtub / cohort Power BI DAX reports bathtub / cohort Failure Tree · graph map · 04 / Analyze · product structure Failure Tree graph map product structure Field Actions · FCO / MTBF · 05 / Act · chart-driven Field Actions FCO / MTBF chart-driven raw work orders full text ticket batch input draft fields candidate accepted records validated prompt fixes iteration PN join basket analysis curated facts read-only fault enumeration graph build failure modes priority list trend alerts KPI Legend primary data async batch data store data flow

Primary Path

  • • Work orders exported read-only from PostgreSQL via Fabric
  • • LLM extractor writes 3-4k tokens per ticket into structured fields
  • • Curated store is the single source for every downstream view

Quality Control

  • • Sampled human review gates convergence of LLM output
  • • Review feedback loops back into prompt fixes, not silent edits
  • • Material master normalized before basket / FPGrowth joins

Downstream Value

  • • Power BI: bathtub curves, cohort views, MTBF trends
  • • Failure tree: data-driven fault enumeration mapped to product structure
  • • Field actions: FCO decisions driven by chart evidence