How Fynd helped an electronics manufacturer replace its legacy MES in 100 days

A fast-growing electronics manufacturer operating SMT production lines, testing, and quality control all within a single facility. Their existing setup used a legacy MES layered on top of SAP - a reliable tool for recording past events, but not designed to react to data in real time or support AI on the shop floor.

At the time of this project, their operations included:

• Multiple SMT production lines, testing stations, and quality checkpoints running 24/7

• A legacy MES built on SAP, lacking native AI or predictive features

• A strict requirement for the new system to run completely on-premises and air-gapped due to facility security concerns

• A tight 100-day deadline to go live, compared to the typical 18-month rollout for similar MES implementations

The challenge: outgrowing a system built for a different era

The legacy MES could track production, but it couldn’t anticipate issues before they happened. As the facility grew, the mismatch between the company’s needs and what the system could deliver became clearer and more pressing.

No room for AI

The old MES couldn’t connect to AI models. Problems with equipment, recurring defects and unusual production patterns were only noticed after the fact - never flagged early enough to prevent downtime or waste.

Slow, costly alternatives

The obvious upgrade path was a commercial MES from a major vendor like Siemens or Rockwell but these came with high licensing costs and typical deployment times of about 18 months. Even then, AI was usually an afterthought, added on rather than integrated.

Operators without real-time support

When issues arose on the line, operators had no quick answers. Every question, from decoding an error to finding the right procedure, meant tracking down a senior technician - slowing down the response time.

Security constraints ruling out cloud-first options

The sensitive nature of the facility meant the new system had to be fully on-premises and air-gapped. This ruled out most modern MES platforms, which are generally cloud-first and rely on external AI services.

The solution: a modular, AI-native MES built from the ground up

Instead of trying to force-fit a commercial platform or an open-source ERP module, the team built a custom MES tailored to the client's production lines - modular by design and built with AI at its core, not as an add-on.

A containerized, microservices architecture

Each MES function from data capture and traceability to dashboards and AI models - runs as an independent service. This made updates straightforward without disrupting production.

A data layer built for both records and real-time signals

PostgreSQL manages transactional data like production orders and component history. InfluxDB captures high-frequency machine and sensor data. Grafana turns all this into live dashboards for supervisors on the floor.

Shop-floor connectivity without heavy custom code

Node-RED connects SMT line controllers and IoT sensors to the system, integrating machines with minimal custom coding.

A 100-day rollout, delivered in three phases

The first 30 days focused on setting up the core platform and linking one pilot line to real production data. By Day 70, the system was live plant-wide, with AI features in test - including predictive maintenance alerts, a defect detection pilot on one camera, and an operator copilot trained on the client’s manuals. The final stretch to Day 100 included shadow-mode testing alongside the old system, a mock recall drill, staff training and full go-live.

AI built into the shop floor, not bolted onto it

An anomaly detection service watches machine data continuously, flagging equipment issues before they cause downtime. Vision-based inspection catches defects during PCB and final assembly stages. And an AI copilot, trained on the client’s manuals and incident logs, helps operators troubleshoot immediately instead of waiting for a technician.

Full ownership, with no recurring license cost

The entire system runs on open-source components, hosted on the client’s servers. There are no per-seat licenses and no vendor release schedules to follow - the client fully owns and can adapt the system as needed.

What this phase covers and what comes next

This initial phase focused on the operational core: production tracking, traceability, quality control, and the first generation of AI features running live in test environments. Predictive maintenance and defect detection will continue to improve as more data flows through the system.

Future phases will expand the AI copilot’s capabilities, extend defect detection to more inspection points and roll the platform out to additional production lines as the facility grows.

Why this approach matters

Manufacturers don’t get stuck for lack of ambition - they get stuck when their systems can’t keep up with what the business demands. This project shows that going AI-native doesn’t have to mean an 18-month commercial rollout or being locked into a vendor’s timeline. When designed right, an open, modular MES can go live in just 100 days, run fully on-premises, and put AI at the heart of the shop floor from day one.

For any manufacturer considering a legacy MES upgrade, the lesson is clear: speed and security don’t have to come at the cost of intelligence. If the system is built for AI from the start not retrofitted later - it can deliver all three.

Curious what this looks like on the ground?

Book a demo