The challenge
Across four plants, line-side QA defects were caught hours after the fact by manual sampling, and component RFQs ran on email threads with no consolidated supplier history. Scrap was eating into margin even on otherwise healthy lines, and procurement could not benchmark new quotes against the last twelve months of pricing.
How we approached it
Ajuni placed Pragna against the MES anomaly stream from each plant and wired Dwight into SAP S/4HANA for vendor history and RFQ orchestration. Edge inference ran on plant-side hardware so the QA loop stayed sub-second, with reconciliation back to the central tenant nightly.
Outcomes in production
- ₹2.4 cr / year of scrap eliminated across 4 plants
- QA anomaly detection moved sub-second on the line
- RFQ cycle benchmarked against 12 months of pricing
- Supplier negotiation prep automated end-to-end
- 9 weeks kickoff to production traffic
Stack & guardrails
Integration & deployment
- SAP S/4HANA integration
- Plant MES connector
- Edge inference · plant hardware
- Llama 3.1 70B · central tenant
- Azure India · production tier
- Power BI mirror
Compliance & audit
- IATF 16949
- ISO 27001
- DPDP Act 2023
- Internal SOX review
Timeline
-
Week 0–2
Plant + ERP mapping
Four-plant MES topology and SAP integration points scoped with platform team.
-
Week 3–4
Edge inference bring-up
Plant-side hardware tuned to keep QA loop under one second.
-
Week 5–7
Shadow QA + RFQ
Agents ran parallel to line QA and procurement; outputs reviewed daily.
-
Week 8–9
Full-plant rollout
Production across 4 plants; audit-trail review with internal SOX team.
