← Back to Platform

AI Assistant

Ask operational questions in plain language. Get grounded, traceable answers from your production data — not estimates, not hallucinations.

What It Is

Conversational Manufacturing Intelligence

The EKAS AI Assistant is a conversational interface to your manufacturing data. Instead of building dashboards or writing SQL queries, operations teams ask questions in plain language and receive grounded answers tied directly to production records.

Every answer comes from your MES or ERP data. Every metric is computed from your confirmed production records using governed definitions. No estimation. No interpolation. No generic model output.

Example Questions

What Operations Teams Ask

What was our worst performing workcenter yesterday?
Which failure modes are recurring on Machine 47?
Show me First Pass Yield by part for the past week
How many hours of downtime did we experience on 2nd shift last month?
Which machines are running below target OEE?
What's driving time variance in stamping this quarter?
Compare downtime between facilities for Q1
Which defect types are trending up this month?
How It Works

Grounded by Design

You ask in plain language

No SQL required. No dashboard configuration. Ask the question the way you would ask a colleague.

Intent is classified and entities are resolved

The platform identifies what you are asking about — which machine, workcenter, time window, or metric — and resolves it against your ISA-95 equipment hierarchy. Prior investigation context is loaded from your conversation history.

Evidence is collected before the answer is formed

Every metric is computed from your confirmed production data. Raw values, record counts, coverage, and data source are assembled into an EvidencePacket alongside the response — before any answer text is generated.

Every number is verified against its source

The platform cross-validates every numerical value in the response against the EvidencePacket collected during the query. A response with numbers that cannot be sourced is not delivered as if it were complete. You receive an honest, structured explanation of what was and was not determinable.

You can drill down or ask follow-ups

Refine the question, change the time window, or explore related metrics in a conversational flow. Your investigation context persists across turns — you do not repeat yourself.

Every query is logged and auditable

Full audit trail of who asked what, when, and what data was accessed — with the complete EvidencePacket preserved. Designed to support IATF 16949 compliance requirements.

Why It's Different

Not a Generic Chatbot

Grounded in your production data

Every answer comes from your MES or ERP. If the data isn't there, the answer isn't there. No estimation.

Deterministic calculation

Metric calculations use governed, versioned definitions applied consistently to your production data. Results are reproducible, auditable, and traceable to source records.

Manufacturing context built in

The system understands ISA-95 hierarchy, ISO 22400-2 metrics, and manufacturing-specific terminology.

Controlled refusal

When EKAS cannot produce a fully evidenced answer, it returns a structured response with a specific reason — time window has no data, entity not resolved, data quality issue detected. You receive an honest structured failure, not a hallucinated answer.

EvidencePacket on every response

Every EKAS response carries an EvidencePacket: the time window queried, the ISA-95 scope, the data source, the record count, the coverage percentage, and the raw values behind every metric. No black-box answers. Every number has a named source.

Full audit trail

Every query is logged with cryptographic hash, catalog version, and data provenance.

Trust Boundaries

What the AI Can and Cannot Do

The AI Can:

  • Understand your question
  • Select the right analytical tool
  • Choose the correct metric
  • Format and explain results
  • Suggest follow-up questions

The AI Cannot:

  • Perform the calculation itself
  • Estimate missing data
  • Interpolate or extrapolate
  • Generate synthetic results
  • Override governed metric logic
The Evidence Standard

EvidencePacket — What Every Response Contains

Every EKAS response carries an EvidencePacket. This is not optional — it is an architectural constraint. A response without a complete EvidencePacket does not reach the user.

time_window
Exact date range of data consumed in this response
scope
ISA-95 nodes covered — site, workcenter, machine, part
grain
Granularity — shift, daily, hourly, or operation level
record_count
Number of source records consumed
coverage_pct
Fraction of requested scope with confirmed data
numerator + denominator
Raw values behind every ratio metric — no black-box percentages
source
Which data view or table was queried

If a required EvidencePacket field cannot be populated, the response is blocked before it reaches the user. This is what 'evidence by contract' means — enforced architecturally on every query, not as a configurable option.

Quality Assurance

Every Release Tested Against Manufacturing-Specific Questions

EKAS maintains a stratified golden query set covering the full range of manufacturing intelligence questions: OEE investigation at every ISA-95 level, downtime root cause analysis, financial cost variance, cross-domain queries combining operational and financial data, document retrieval, and structured failure scenarios. Every release is evaluated using the RAGAS framework — measuring faithfulness (answer grounded in source data), relevancy (answer addresses the question), and context precision (correct data retrieved). Releases that regress on any dimension do not reach production.

See the AI Assistant in Action

Bring an operational question. We'll show you how EKAS answers it with grounded production data.