Correcting Tag Errors with Context - Fixing What OCR Missed

Challenge

The client’s workflow involved processing scanned vendor P&IDs, many of which contained:

• Low-contrast tag numbers
• Inconsistent formatting (e.g., A1-FT-1O3 instead of A1-FT-103)
• Minor OCR distortions that made automated post-processing difficult

These small errors broke tag associations in downstream MTOs and created costly rework. Manual correction was slow, and internal teams lacked a structured way to decide:

“Is this tag 1O3, 103, or 1D3 — and how sure are we?”

Solution

Storm Consulting enabled the OpenAI-enhanced reasoning layer in eAI, allowing:

• Automated suggestions for corrected tag numbers
• Use of surrounding context (nearby tags, loop numbers) to infer likely values
• Human review of suggestions before committing changes

The prompt logic used only anonymized, cleaned text, and no actual drawings or vendor-specific data were sent to OpenAI.

Implementation Highlights

• Applied a “tag correction prompt” with 2–3 nearby tag samples
• Limited outputs to tag format rules (e.g., XX-FT-123, 41-PT-255)
• Suggestions integrated directly into the annotation UI
• Engineers could accept, reject, or edit the correction
• Every change was logged with before/after states for traceability

Results

“I didn’t expect a language model to understand P&ID tags — but it consistently suggested the right corrections. That changed everything.” — QA Engineer, Client Team

💡 Why It Worked

• Engineers trusted the prompt-driven suggestions, since they could see the logic behind each correction
• Tag formatting rules and nearby references anchored OpenAI’s responses
• The system never auto-applied changes — validation stayed in the loop
• Suggestions were contextual, not just spelling corrections

Share

Like this? Share it with your network!

• Share on Facebook
• Share on Twitter
• Share on LinkedIn