labs@atheron:~/industries/epc/ai$ train agent --grounded

AI for EPC engineering

Search specs, drawings and vendor data, often on premise.

Engineering projects run on more documents than anyone can read. AI earns its place in finding the answer and the page it came from, checking one document against another, and preparing the routine checks an engineer then confirms. Because client and vendor documents are confidential, the model often runs on premise.

Start a project See the example projects

// what matters here

What is different in this industry.

Confidential data stays inside

An open-source model on a GPU server in your data centre, or in your own cloud tenant, so specifications and vendor data never go to an outside provider.

Every answer cites its source

The assistant points to the document and page behind each answer, so an engineer can check it before relying on it.

Checks, not decisions

It prepares MTO checks and flags inconsistencies between documents; an engineer confirms or rejects each one, and the record shows who did.

Measured before it ships

An evaluation set built from your engineers' real questions is run on every change, and the score is a gate.

// example projects

Priced examples for this service.

Each one is hypothetical, labelled as such, and priced live from our rate card. Open any of them in the estimator and make it yours.

Example projectWeb appAIOpen-source model

On-premise spec search assistant

Hypothetical. Not a client, not a result.

// The problem

Engineers search thousands of specifications, drawings and vendor documents to answer one question, and contract terms mean those documents cannot go to an outside model.

// What we would build

An assistant on an open-source model running on a GPU server in your own data centre that answers from specifications, drawings and vendor data with the page it came from, prepares MTO checks and flags inconsistencies between documents for an engineer to check.

// What is in it

Specifications, datasheets and vendor documents indexed and kept in sync
Answers that cite the document and page they came from
MTO checks drafted against the specs, for an engineer to check
Flags where two documents disagree on a value
An evaluation set from real questions, run before every change
An open-source model on a GPU server you own, with nothing sent outside

// Stack

Llama or Qwen (open-source)
vLLM
PostgreSQL with pgvector
Next.js
Lenovo GPU server on premise

// estimate

Build: ≈ US$148,000 to US$225,000, delivered within 27 weeksCAD 210,600 to 320,600
Hosting: ≈ US$176 a monthCAD 250 a month
Support: ≈ US$2,500 a monthCAD 3,565 a month
Hardware: ≈ US$47,800 to US$197,000CAD 68,000 to 280,800

Prices in your currency are estimates from today's Bank of Canada rate. All invoicing is in CAD or USD.

AI route: Open-source model
Model running cost: ≈ US$176 a monthCAD 250 a month

Timeline by milestone

Discovery: 2.9 to 3.4 weeks
Specification and evaluation plan: 0.4 weeks
Design approved: 2.4 to 3.9 weeks
Core features: 1.4 to 2.4 weeks
Full build: 0.9 to 1.4 weeks
Working pilot: 5.9 to 10.4 weeks
Testing and fixes: 1.4 to 2.4 weeks
Launch: 0.9 to 1.4 weeks
Production: 0.9 to 1.4 weeks

Outside our hands, and added to the calendar

Hardware delivery, after it is ordered: 2 to 6 weeks
Client IT security review and vendor onboarding: 4 to 8 weeks

How it is paid

Deposit 20%: CAD 42,060 to 64,060
Discovery 1.6%: CAD 3,364.80 to 5,124.80
Specification and evaluation plan 11%: CAD 23,133 to 35,233
Design approved 1.6%: CAD 3,364.80 to 5,124.80
Core features 4.9%: CAD 10,304.70 to 15,694.70
Full build 3.2%: CAD 6,729.60 to 10,249.60
Working pilot 33.1%: CAD 69,609.30 to 106,019.30
Testing and fixes 1.6%: CAD 3,364.80 to 5,124.80
Launch 1.6%: CAD 3,364.80 to 5,124.80
Production 11.4%: CAD 23,974.20 to 36,514.20
Holdback, 30 days after launch (10%): CAD 21,030 to 32,030
GPU time for training and testing, at cost: CAD 300. Billed up front, at cost, outside the milestones
GPU server for AI, for the model: CAD 68,000 to 280,800. Billed up front, at cost, outside the milestones

Start from this example

// questions

Questions we get about this.

Can it read our drawings as well as text documents?

It reads text from drawings and title blocks, and the attributes your design tools export. Reading geometry from drawings is a larger project we would scope separately.

Which model route fits engineering data?

Usually an open-source model on premise or in your tenant, because of confidentiality clauses. A frontier model is quicker to start where the documents are not sensitive. Each route is priced on its own.

How do you keep it from inventing values?

It answers only from retrieved documents, cites them, and says when it cannot find an answer. The evaluation set includes questions designed to catch invented values.

Back to the industry: Engineering and construction (EPC/EPCM)About the service: AI agents and models

// next

Not quite your project?

Tell us what you have in mind. We will come back to you with a range and the questions that would narrow it. Or book a call and talk it through.

Start a project Book a call