MAINFRAME MODERNIZATION

Reinventing Mainframe Systems with Enterprise Modernization.

Industry

Manufacturing — Discrete & Process (shopfloor, suppliers, SCM, ERP)

Technology

Edge gateways (OPC-UA / MQTT), Industrial IoT firmware, Kubernetes (edge + cloud), Time-series DB (InfluxDB/Prometheus), Vector DB for embeddings, Open-source LLMs (fine-tuned), FastAPI microservices, Kafka (event bus), ERP/SCM connectors (REST / EDI), Grafana dashboards, CI/CD and IaC.

Overview

A large manufacturing plant wanted continuous, real-time visibility from devices on the shopfloor through to supply-chain systems, automated supplier visibility during MRP runs, and ML/LLM-driven procurement intelligence — all while meeting a strict availability target (>98%) for production-critical equipment. Sorim.AI delivered an edge-first platform that ingests device telemetry, synchronizes plant + SCM + enterprise data, and applies open-source LLMs to surface buying patterns and negotiation insights for procurement teams.

Problem statement

"Embed software to connect external device data and continuously monitor devices remotely to ensure availability above 98%.

Automate 2 & 3-tier supplier visibility during MRP runs.

Automate device data, plant data and enterprise data to ensure plant and operation availability is high."

Intermittent device connectivity and long detection-to-repair times.
Lack of automated supplier visibility during MRP — manual follow-ups and frequent stockouts.
Siloed plant and enterprise data that prevented predictive insights and rapid troubleshooting.
Need to avoid heavy changes to existing PLCs and ERP while still modernizing data flow.

Challenges

Intermittent telemetry & legacy protocols: Many machines spoke OPC-UA/serial and produced noisy telemetry.
Latency & reliability: Central cloud polling caused delayed alerts and missed events during network issues.
Supplier visibility: 2nd/3rd tier suppliers weren’t part of the MRP loop; manual checks caused BOM shortages.
Data fusion: Merging high-velocity device streams with ERP/SCM transaction data in near-real time.
Actionable intelligence: Transforming raw telemetry + supply signals into procurement negotiation insights and operator-actionable alerts.

Solution Overview

1) Edge-first data fabric

Deployed lightweight edge agents on shopfloor gateways that normalize OPC-UA / MQTT / Modbus data, perform on-device anomaly detection (rolling windows + lightweight models), and buffer telemetry during network outages. Agents publish compressed event streams to a resilient Kafka cluster (edge → cloud) for ordering and replay.

2) Real-time observability & control

Time-series platform (Prometheus/InfluxDB) + Grafana dashboards for OEE, MTTR, and device health; alerting with escalation rules and runbook links. Remote command channel for safe, authenticated device interactions (read / soft reset / config push).

3) MRP & supplier orchestration

Built connectors to ERP/SCM (REST / EDI) to ingest purchase orders, lead times, and supplier commitments. During MRP run the platform automatically queries 2nd/3rd tier supplier statuses, lead-time deviations, and suggested alternate suppliers; results appear inside the planner UI as ranked recommendations.

4) LLM + Pattern Intelligence for procurement

Aggregated historical SCM, purchase, and regional market data into a vector DB; fine-tuned an open-source LLM to answer procurement queries like “best region to source X at current demand” and to generate negotiation talking points. The LLM produced supply-risk narratives and price trend summaries that procurement teams used during supplier negotiations.

5) Platform resiliency & governance

Role-based access control (operators, planners, procurement, execs), audit trail for every remote action, and Canary deployments for edge agent updates via CI/CD.