Skip to content
AtheronLABS

You're visiting from the United States. Prices are shown in US dollars. Not right?

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

AI for clean energy

Fleet monitoring and an O&M assistant for technicians.

A renewable fleet produces plenty of data and not much time to read it. The useful AI here is narrow: flag the sites producing less than they should, and help a technician find how the same fault was fixed last time.

// what matters here

What is different in this industry.

01

Expected against actual

Each site's output is compared with what irradiance and wind say it should be, so a soiled array or a derated inverter shows up in days, not at month end.

02

Answers with their source

The assistant answers from your manuals and work orders and cites the page or the order, so a technician can check it on the way to site.

03

Private by default

An open-source model served in your own cloud account keeps maintenance records and contracts out of anyone else's systems.

04

Measured before each change

An evaluation set of your technicians' real questions is run before every model or prompt change, and the result is kept.

// 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

Renewable fleet monitoring

Hypothetical. Not a client, not a result.

// The problem

An independent power producer watches its sites in each maker's portal, finds underperformance at month end, and its technicians search old work orders for how a fault was fixed.

// What we would build

One monitoring dashboard that reads every site through the makers' APIs, compares output with what the weather says it should be, and an assistant on an open-source model hosted in the producer's cloud that answers from manuals and past work orders, citing them.

// What is in it

  • Inverters, turbines and battery systems read through their makers' APIs
  • Expected output from irradiance and wind data, with gaps flagged daily
  • Alarms grouped by site and cause, not one email per fault
  • An assistant that answers from manuals and work orders and cites them
  • An open-source model served on GPUs in your own cloud account

// Stack

  • Next.js
  • PostgreSQL with TimescaleDB
  • Python
  • Llama or Mistral (open-source)
  • vLLM
  • DigitalOcean GPU

// estimate

Build
≈ US$127,000 to US$194,000, delivered within 23 weeksCAD 180,900 to 276,400
Hosting
≈ US$4,530 a monthCAD 6,450 a month
Support
≈ US$2,500 a monthCAD 3,565 a month

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$4,190 a monthCAD 5,960 a month

Timeline by milestone

Discovery
2.9 to 3.4 weeks
Specification and evaluation plan
0.4 weeks
Design approved
2.9 to 5.4 weeks
Core features
1.4 to 1.9 weeks
Full build
0.9 to 1.4 weeks
Working pilot
3.9 to 6.4 weeks
Testing and fixes
0.9 to 1.9 weeks
Launch
0.9 to 1.4 weeks
Production
0.9 to 1.4 weeks

Outside our hands, and added to the calendar

API keys from each maker or the SCADA owner
1 to 4 weeks

How it is paid

Deposit 20%
CAD 36,120 to 55,220
Discovery 2%
CAD 3,612 to 5,522
Specification and evaluation plan 10.3%
CAD 18,601.80 to 28,438.30
Design approved 2%
CAD 3,612 to 5,522
Core features 6%
CAD 10,836 to 16,566
Full build 4%
CAD 7,224 to 11,044
Working pilot 31.1%
CAD 56,166.60 to 85,867.10
Testing and fixes 2%
CAD 3,612 to 5,522
Launch 2%
CAD 3,612 to 5,522
Production 10.6%
CAD 19,143.60 to 29,266.60
Holdback, 30 days after launch (10%)
CAD 18,060 to 27,610
GPU time for training and testing, at cost
CAD 300. Billed up front, at cost, outside the milestones

// questions

Questions we get about this.

Do we need our own weather data?

No. Satellite irradiance and weather services cover most sites; your own met stations improve the comparison where you have them.

Why an open-source model for the assistant?

Maintenance records and service contracts are commercial. An open-source model in your cloud keeps them private; a frontier model is cheaper to start if they are not sensitive.

Can it predict failures?

It flags underperformance and patterns in alarms. Failure prediction needs enough failure history to measure against, which we check before promising anything.

// 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.

AI for solar, wind and battery operations and maintenance | Atheron Network Labs