Skip to content
HJ ConsultHJ Consult
esc
Start a brief
Services
Work
Field notes
About
Brand
Norsk
Englishen
Switch mode: Nordlys
Switch mode: Systems
↑↓ select, ↵ open, esc close⌘K
Mode

Services, Norway and remote

What we build, we also run

AI, custom software and cloud from HJ Consult, with one contact from brief to operations. Below: what that means in practice, what we use, and answers to the questions people actually ask.

Three pillars 01–0316 capabilities in 7 categories

Three pillars

Short answer first, details after

  1. 01AI solutions

    AI and language models

    Typical stack
    • Mistral (EU endpoint)
    • Claude
    • Vercel AI SDK
    • pgvector
    • Supabase
    Short answer

    HJ Consult builds AI solutions for businesses: LLM agents, search over your own documents (RAG) and automation of manual routines. We run Mistral on EU servers (and Claude where it fits) and put everything into production with monitoring, not as a demo.

    What we do
    • Agents that do work inside your systems, with clear limits and an audit trail.
    • RAG search that answers from your documents, with sources instead of guesses.
    • Automation of routines: classification, form-filling, follow-ups.
    • Model training on rented GPU capacity when off-the-shelf models fall short.
  2. 02Custom software

    Custom software

    Typical stack
    • TypeScript
    • React
    • Next.js
    • React Native
    • Node.js
    • Python
    • PostgreSQL
    Short answer

    HJ Consult develops custom software from idea to finished product: web, mobile and backend in TypeScript, React, Next.js and PostgreSQL. You own the code from the first commit, and we build it so it can keep changing after delivery.

    What we do
    • Web and mobile apps with design and front-end that can stand being looked at.
    • Backends, APIs and integrations: payments, messaging, locks, heaters.
    • Admin panels, sign-in and multi-tenant setups on Supabase.
    • Firmware and control when your product has a physical side.
  3. 03Cloud & operations

    Cloud, architecture and operations

    Typical stack
    • Vercel
    • Supabase
    • AWS
    • Google Cloud
    • Docker
    • GitHub Actions
    Short answer

    HJ Consult sets up, runs and monitors cloud platforms on Vercel, Supabase, AWS and Google Cloud. We automate deployments with GitHub Actions, keep costs visible and take the call when something beeps.

    What we do
    • Architecture that fits your size today and next year.
    • CI/CD, environments and secrets set up properly from the start.
    • Monitoring, alerting and logs you can actually read.
    • Cost control: you see what costs what before the invoice does.

How we work together

One contact, fixed phases and a demo every week

No big reveal at the end, and no hours that quietly disappear. A typical engagement looks like this:

  1. 01How we work together

    Brief

    We start with a brief, not a quote. You say what you want to achieve; we draw up architecture, phases, risks and an honest estimate.

    You getA document you can take to the board.
  2. 02How we work together

    Fixed-scope phases

    The work is split into phases with agreed scope and price. Want to change course? We do it between phases, never in a footnote.

    You getPredictable cost and a clear order of work.
  3. 03How we work together

    Weekly demo

    Every week you see something that works. You click through it yourself, we take notes, and next week it is better.

    You getA product you recognise before it is finished.
  4. 04How we work together

    Handover or operations

    The code lives in your repository with documentation and a runbook. We are happy to keep running it, but you can always take it in-house.

    You getA system without lock-in.
WhereNorway and remote

We are based in Bergen and work remotely, across the Nordics and the EU. Meetings happen on video, and in person when it actually makes a difference.

Capacity
Status
Ask and we’ll tell you
Start
We agree a date by email
Checked
September 13, 2026

We don’t run a countdown and we don’t hold seats. When it is full it says full. The date is the last time we looked at the calendar.

Last time round

Four sentences you would rather not hear again

It is rarely bad faith. These sentences usually mean that something was agreed, or not agreed, in week one. Here is what they tend to cover for, and what we do differently.

  1. “It turned out to be a bit more work than we thought.”

    What it means

    The work was always there. It just was not in the quote, because nobody had drawn the system before the number was set.

    With us

    Our estimate is often the highest in the pile, because the hard parts are in it. That is a conversation we would rather have before you sign than after.

  2. “We’re in the final stretch now.”

    What it means

    You have not seen anything for six weeks, and the final stretch has no date. Usually that means there is nothing to show.

    With us

    A week with no progress is visible too. It sits in the same calendar as the good ones.

  3. “The pilot went well. We’ll look at production next year.”

    What it means

    The pilot ran on tidy data with somebody watching it. The hard part (errors, access, cost, who finds out when the model gets it wrong) was not in it.

    With us

    We ask who notices the mistake before we ask what the model should do. If the answer is nobody, it is not a pilot. It is a demo.

  4. “We’ll send over a zip when we’re done.”

    What it means

    The code exists. The history, the environment variables and the three manual steps nobody wrote down do not. The next developer starts by guessing.

    With us

    The runbook gets written while we are in there, not at the end. It is the only way those three steps make it out.

We are no exception. Our own AI review died on six of twenty-four Norwegian runs, and it looked like an unstable model. It was an empty list in a schema we wrote ourselves. The difference is where the mistake turns up: in a demo you are watching rather than an invoice you open.

Read the bug report

Range

What we have built before

An app for an energy company, and the website for a construction firm. A customer-service agent that answers sauna guests around the clock and learns from the conversations. Search that reads a company’s own documents and answers from them. Automation that takes the routine work nobody wants. Systems that read markets in real time. Data collection at scale, payment flows, and control of physical equipment standing out in the weather.

Some of it is our own. Much of it was built for other people, and most of that we cannot show: as a rule it is exactly those clients who would rather not have their name on a website. But that is where we learned what holds up in production, and what only holds up in a demo.

The whole catalogue

Chances are we have built something like it

General claims are cheap, so every line names the technology behind it. Some are running for us today; the rest we build the same way.

16 capabilities in 7 categories

01The whole catalogue

AI solutions

Language models, agents and automation that actually reach production.

  1. AI solutions

    We build LLM agents, retrieval-augmented generation and automation that ship to production, with data that stays in the EU.

    • Mistral (EU endpoint)
    • Claude
    • Vercel AI SDK
    • pgvector
    • Supabase
02The whole catalogue

Custom software

Web, mobile and backend, built to last and to change.

  1. Custom software

    From idea to finished product: web, mobile and backend built to last, and to change when the world does.

    • TypeScript
    • React
    • Next.js
    • React Native
    • Node.js
    • Python
    • PostgreSQL
    See HeatBooking
03The whole catalogue

Cloud & operations

Platforms that scale without the bill running away.

  1. Cloud & operations

    We set up, run and monitor cloud platforms that scale with you without the bill running away.

    • Vercel
    • Supabase
    • AWS
    • Google Cloud
    • Docker
    • GitHub Actions
04The whole catalogue

Payments

Whichever payment provider you need, integrated on request.

  1. Payment integrations

    We integrate whichever payment provider you need on request, with security in place from the first transaction.

    • Vipps
    • MobilePay
    • Stripe
    • Klarna (via Stripe)
    • Neonomics (open banking, PSD2)
    See HeatBooking
05The whole catalogue

Hardware & control

Anything that fits an industrial PLC, plus the apps that control it.

  1. PLC & firmware

    Anything that fits an industrial PLC: we write firmware from scratch, like the control systems for saunas and heating.

    • Arduino Opta
    • Firmware from scratch
    • Sauna and heat control
    See Vestfjord Sauna
  2. Sensors & heater control

    Precise temperature sensing with PT100 through LKM264 transmitters, and direct API control of HUUM sauna heaters.

    • PT100
    • LKM264 transmitters
    • HUUM heater API
    See Vestfjord Sauna
  3. Access control

    Smart locks and access tied to the booking, with no key boxes and no weekend phone calls.

    • igloohome
    • Inlet
  4. Remote control & dashboards

    Apps and dashboards that control the equipment from anywhere, by phone or from the control room.

    • React Native
    • Next.js
    • Real-time dashboards
    See Vestfjord Sauna
06The whole catalogue

Data & AI

From raw data to models, trading and dashboards you can trust.

  1. Algorithmic trading

    Trading systems that decide on data rather than gut feeling, with backtesting and risk controls built in.

    • Python
    • Backtesting pipelines
    • Exchange and broker APIs
  2. Machine learning & GPU training

    We train models on burst GPU capacity from Verda and others. You pay for the hours, not the hardware.

    • Verda
    • Burst GPU capacity
    • PyTorch
  3. LLM agents, RAG & automation

    Agents that do the work, retrieval that finds the answer and automation that removes the boring parts.

    • Mistral (EU endpoint)
    • Claude
    • RAG
    • pgvector
    • Model Context Protocol
    See Vestfjord Sauna
  4. Analytics, data pipelines & dashboards

    From raw data to numbers you can trust: pipelines, models and dashboards people actually use.

    • PostgreSQL
    • Python
    • Supabase
    • Dashboards
07The whole catalogue

Platforms

Booking, messaging, sign-in and operations: the building blocks behind a product.

  1. Messaging & notifications

    SMS, email and push that get through, with delivery tracking and templates you control.

    • ClickSend SMS
    • Resend
    • Web push
  2. Booking, memberships & scheduling

    Complete booking systems with memberships, capacity and payments, built and run by us.

    • HeatBooking
    • Stripe
    • Vipps
    See HeatBooking
  3. Auth, admin panels & multi-tenant SaaS

    Secure sign-in, admin panels and multi-tenant setups on Supabase, ready for the next customer from day one.

    • Supabase Auth
    • Row Level Security
    • PostgreSQL
  4. Monitoring, DevOps & CI/CD

    Automated deployments, monitoring and alerting, so you know things work before customers tell you.

    • GitHub Actions
    • Vercel
    • Docker
    • Uptime monitoring

Questions we get a lot

Honest answers to honest questions

How do we start?

With a brief. Write one sentence about what you want to build in the Living Brief, or send an email. We reply with questions, not a quote. The first conversation carries no obligation.

What does it cost?

It depends on the scope, which is why we start with the brief. From there we agree scope and price per phase, so you know what the next step costs before you say yes to it.

Where does our data end up?

In the EU. Language models run against Mistral’s EU endpoint, the database lives with Supabase in Stockholm, and server functions run in Vercel’s Stockholm region. That includes everything you type into the Living Brief.

Do we own the code?

We normally work with the code in your repository from the start, together with documentation and a runbook. The goal is that you can carry on without us.

Can you work with hardware and control systems?

Yes. We have written firmware from scratch for Arduino Opta, read PT100 sensors through LKM264 transmitters, controlled HUUM heaters over API and tied igloohome and Inlet locks to bookings. Software that meets the physical world is our favourite kind.

How big or small are the projects you take on?

Anything from a few weeks’ integration to a whole product we build and run. We are small, so we say no when we are not the right fit, and yes when we are.

Do you run it afterwards?

Gladly. Everything we build, we can also monitor, update and maintain. If your team wants to take over, you get a tidy handover instead.

Where are you, and how do we reach you?

Bergen, working remotely. Send an email or start a brief and we will book a call that suits you.

Next step

Tell us what you want to build

One sentence is enough. The brief takes it from there.