AI-Powered Manufacturing

Your Shop Floor
Speaks English Now

PulseMQ's built-in AI agent uses Claude on AWS Bedrock to give every operator, supervisor, and engineer a natural-language interface to your entire MES. Ask questions, dispatch jobs, troubleshoot issues — no training manual required.

Anthropic Claude
AWS Bedrock
Self-Learning
PulseMQ AI Agent — Extruder Line 02
Online
Why is OEE dropping on the extruder?
GetMachineStatus
I checked MACH-EXT02. OEE dropped from 82% to 61% over the last 2 hours. The main culprit is Performance — line speed fell from 12.4 m/min to 8.1 m/min at 14:23.

The temperature is also drifting: zone 3 is at 192°C vs setpoint 185°C. That's likely causing the speed reduction (thermal protection). Want me to check the recipe parameters?
Yes, and what material lot are we running?
GetJobDetails • GetRecipe
Current job: JOB-20260312-004 (WO-2026-0082)
Recipe: PVC-COMPOUND-R101 v2.1 — temp setpoints are Z1: 180°C, Z2: 183°C, Z3: 185°C
Material: Lot PVC-20260311-019 (loaded at 06:12 today, 1,847 kg remaining)

Zone 3 is 7°C over setpoint. This lot has a different melt index than the previous one — the PLC may need a recipe adjustment. Want me to flag this for the supervisor or create a maintenance ticket?
Create a maintenance ticket for the heater band check on zone 3
CreateMaintenanceTask
Done. Created maintenance task MT-0312-007:
Type: Corrective
Machine: MACH-EXT02 — Zone 3 Heater Band
Priority: High
Description: Zone 3 temp drifting +7°C above setpoint. Possible heater band degradation. Check PID tuning and heater element resistance.

I've also linked it to the current job and material lot for traceability.

What the Agent Can Do

Not a chatbot with canned responses. A real AI agent with tool access to your live production data, machine configs, recipes, and work orders.

Live OEE Analysis

Ask "why is OEE low?" and get a breakdown of availability, performance, and quality with root cause analysis tied to real-time machine signals.

Job & Work Order Lookup

Query active jobs, scheduled work orders, output records, and material lots across any machine. "What's running on the mixer?" works.

Maintenance & Alarms

Create maintenance tickets, check PM counters, and investigate alarms. The agent understands which machine, which component, and links everything to the current job.

Material Traceability

Track material lots from receiving through production. "What lot is loaded?" and "Where did lot PVC-001 go?" both work with full genealogy.

Recipe & Config Guidance

Get help configuring machines, understanding ISA-88 recipes, or troubleshooting ISA-106 state transitions. The agent knows your data model inside and out.

Fuzzy Search

Don't remember the exact machine name? Say "the saw" or "mixer 1" and the agent fuzzy-matches it to the right machine, product, or recipe.

Not Static. Self-Learning.

Traditional MES assistants are keyword lookups with canned responses. PulseMQ's agent understands context, reasons about your production data, and improves with every interaction.

1
Operator asks a question
"Why did we reject 3 rolls on the last job?"
2
Agent reasons & calls tools
Queries job outputs, state history, material lots, temperature signals
3
Correlates across data sources
Finds temperature spike during state transition + new material lot loaded
4
Delivers actionable answer
Root cause + recommendation + creates follow-up task if requested
5
Retains context for next question
Conversation history builds domain understanding over time

Deep Domain Knowledge

The agent is seeded with complete knowledge of ISA-88, ISA-95, ISA-106, Sparkplug B, OEE calculations, recipe management, and PulseMQ's data model. It doesn't just parrot — it understands.

Tool-Use Architecture

Not a prompt-and-pray chatbot. The agent has structured tools — GetMachineStatus, GetJobDetails, SearchProducts, CreateMaintenanceTask — that query live DynamoDB data through your API.

Role-Aware Responses

Operators get quick status answers. Supervisors get production summaries and dispatch options. Admins get configuration guidance. Same agent, different depth based on who's asking.

Continuous Improvement

As your production data grows, the agent gets better context. New machine configurations, recipe changes, and historical patterns all enrich future answers. The system learns your factory.

Architecture Stack

Enterprise-grade AI infrastructure. No self-hosted models, no GPU servers, no prompt engineering on your part.

Layer 1
PulseMQ Dashboard (React)
Operator types natural-language question in the chat panel. Context (machine, role, shift) sent automatically.
Layer 2
MES Agent Lambda (Python)
Builds context-aware system prompt with deep ISA-88/95/106, Sparkplug B, and OEE domain knowledge. Manages conversation state and tool orchestration.
Layer 3
AWS Bedrock — Claude (Anthropic)
Foundation model inference. Claude reasons about the question, decides which tools to call, and synthesizes production data into human-readable answers.
Layer 4
Tool Execution (DynamoDB + API Gateway)
Agent calls structured tools: GetMachineStatus, GetJobDetails, SearchProducts, GetRecipe, CreateMaintenanceTask, DispatchWorkOrder. All hit live production data.
Layer 5
Production Data (Machines, Jobs, Materials, Recipes)
DynamoDB tables: Machines, MachineJobs, PivotedMachineData, CustomerRecipes. Real-time data from MQTT + Sparkplug B. The single source of truth.
0
GPU Servers to Manage
AWS Bedrock = serverless AI
<3s
Average Response Time
Including tool calls + reasoning
100%
Data Stays in Your AWS Account
No external API calls or data sharing

Stop Reading Manuals.
Start Asking Questions.

PulseMQ's AI agent turns every operator into a power user. Natural language, real-time data, zero training curve.

See the AI Agent Live