Flagship Build
What your engineers could ask
An AI assistant built on your plant's own documentation: PLC programs, submittals, manuals, and the knowledge in your senior engineers' heads. Every answer cites its sources. When the documents are silent, it says so.
This is Otoniq's flagship system, one of the things we build and integrate as part of our services.
Inputs
What we ingest
If your engineers use it to run the plant, it belongs in the document set.
PLC programs
Siemens TIA Portal exports and Rockwell L5X files, parsed down to blocks, tags, and cross-references.
IO lists and tag databases
Signal lists and tag databases become the index that ties every answer back to real equipment.
Electrical submittals
Motor protection settings, panel schedules, and instrument datasheets: the details that decide a fault call.
O&M manuals
Operating envelopes, maintenance intervals, and troubleshooting tables from vendor manuals.
P&IDs
Process and instrumentation diagrams, so the assistant can walk a loop the way an engineer would.
Commissioning and handover reports
Setpoint changes, workarounds, and as-left settings that never made it into the manuals.
Scanned and legacy PDFs
Older documentation gets OCR'd and indexed. Age is not a reason to leave it out.
Captured tribal knowledge
Structured interviews with senior engineers, turned into citable documentation before it walks out the door.
Worked Example
Every answer shows its sources
Ask in plain language. Get an answer grounded in your own documents, with the citations to check it.
Level transmitter LT-204 reads 100% but the sight glass shows half full. What does the loop look like?
LT-204 is a 4–20 mA loop on Tank T-200, scaled in FB_Level_Scaling. The datasheet lists a calibrated range of 0–2.5 m, but the scaling block maps the input to 0–1.25 m — consistent with the transmitter being replaced by a shorter-range unit without the PLC scaling being updated.
Check the as-left calibration sheet against the HI_RANGE parameter in FB_Level_Scaling. A half-full tank reading 100% matches a 2:1 range mismatch exactly.
How many spare mechanical seals do we hold for the T-200 transfer pumps?
That's not in your documentation — I don't guess. The document set covers the seal type and the replacement procedure, but no stores or inventory records were ingested.
▸ Refuses when your documents are silent.
Illustrative exchange. Document names and tags are synthetic.
By Design
It refuses to guess
A wrong answer in a plant costs more than no answer. An invented torque value or a guessed interlock isn't a chat mistake. It's a work order built on fiction. So when your documents don't contain the answer, the assistant says exactly that, and tells you which document would need to exist for it to answer.
Every refusal is information. Over time, the questions it can't answer become your documentation-gaps list: a map of what to capture next, ranked by what your engineers actually ask.
Part of the Ingest
Automated PLC documentation
PLC exports get fully documented, not only indexed for search. Blocks, routines, and tags come out as readable, cross-referenced documentation your engineers can review on its own, before anyone asks the assistant a single question.
How It Starts
The shape of an engagement
Nothing here drags on for quarters. Four steps, and you decide at the end.
Discovery call
Thirty minutes on your plant, your documentation, and whether this is worth doing. We talk through the work, not through a slide deck.
Sample ingest
You hand over a bounded slice of documentation: one area, one line, one system. We build the assistant on it.
Your engineers query it
The people who know the plant best try to break it. They judge the answers against what they know.
You decide
If the answers hold up, we scope the full rollout. If they don't, you've lost little and learned where your documentation stands.
Next Step
See it on your own documentation
Thirty minutes to find out whether your document set can carry an assistant like this, and what it would take.