A research-backed analysis of 7 LLM-agent opportunities across AFMW's 140-year-old Ashland, PA foundry — with verified dollar-figure savings calibrated to confirmed revenue of $31.4M and ~175 employees. All automations use Large Language Models; Computer Vision is used only for visual quality inspection.
Every automation in this analysis is delivered by a Large Language Model agent — reading documents, reasoning over data, and writing outputs for human review. The sole exception is quality inspection, where Computer Vision is used for its inherently visual task. No IoT sensor platforms, no trained ML models, no APS scheduling engines are required.
| Process | Current State | With LLM Agent | Tech | Impact |
|---|---|---|---|---|
| RFQ & Quoting | Manual email/web form; 2–4 hrs per quote | LLM reads RFQ email, extracts specs, drafts quote for estimator review | LLM | 60% time reduction |
| Quality Inspection | Manual visual inspection; ~8% scrap/rework rate | Computer vision detects surface defects; LLM writes non-conformance report | CV + LLM | 30–50% defect reduction |
| Customer Service | Phone/email only; manual MTR and cert delivery | LLM reads inbound emails, retrieves documents, writes complete responses | LLM | 80% of inquiries automated |
| Production Scheduling | Manual scheduling; 'disorganized environment' (Glassdoor) | LLM reads job queue, writes prioritized daily schedule and shift notes | LLM | OTD improvement; planner time freed |
| Process Parameters | Engineer judgment on pour temp, risers, gating for each new job | LLM reads historical job records, writes parameter recommendations | LLM | ~2% yield improvement |
| Procurement | Manual supplier emails, PO drafting, invoice review | LLM drafts POs, writes supplier emails, flags invoice discrepancies | LLM | 0.5 FTE admin time saved |
| HR & Recruiting | Manual resume review; 13 open roles; high turnover | LLM screens resumes, ranks candidates, writes onboarding plans | LLM | 40% faster time-to-fill |
LLM agents that automate AFMW's manual RFQ email workflow, draft quotes, and handle routine customer inquiries end-to-end
Computer vision for defect detection on castings, plus LLM agents for production scheduling, shift reporting, and non-conformance documentation
LLM agents that monitor alloy prices, draft purchase orders, score suppliers, and write process improvement recommendations from existing casting data
LLM agents for recruiting, onboarding, compliance documentation, and accounts payable — addressing AFMW's confirmed high-turnover environment
All savings are generated by LLM agents reading documents, reasoning over structured data, and writing outputs for human review. Computer Vision is used only for casting defect detection. The optimistic total has been discounted by 0.7× to account for implementation friction. AFMW does not publicly disclose financials; all figures are independent estimates.
LLM RFQ automation + LLM customer service + CV quality inspection pilot
LLM production scheduling + LLM process parameter recommendations
LLM HR automation + LLM AP and compliance documentation