Services

AI solutions for business

Most companies do not need a model of their own. What they need is AI applied to the one place that eats staff time. Solv looks at how you work first, says where AI helps and where it does not belong, and only then builds.

What we do

Internal knowledge chatbot

Your processes, documents and handbooks loaded into an assistant staff can question. Answers come back with the source cited, instead of someone digging through a shared drive.

LLM features in existing products

AI added to software or a website you already run. Summarising, classifying, drafting, answering customers. No rebuild required.

Workflow automation agents

AI running a multi-step task end to end, from reading an order to entering data, calling APIs and processing paperwork. Every step is logged so you can check the work.

Data analysis and reporting

Scattered data pulled together into reports and readings. F68 does exactly this with price data, financial statements and market news to build investment reports.

Models and infrastructure

We pick the model to fit how sensitive your data is and what the budget allows. Not whichever one is being talked about. The application around the model uses the same stack as our other work.

Commercial models via API

The strongest models available, with no servers to run. Each vendor is strong at different things and pricing varies widely, so we test on your own data before choosing. Data does leave your network, so we limit what gets sent and document it at handover.

  • OpenAI
  • Anthropic
  • Gemini
  • DeepSeek

Self-hosted open models

The model runs on your own servers and nothing leaves your system. Infrastructure costs more and quality sits below the commercial models. Still, it is the only option when the data cannot go out.

  • Llama
  • DeepSeek
  • Ollama
  • Docker
  • Private servers

The application around the model

The model is only one part. The rest is where documents live, how relevant passages are found, how users are permissioned, and the conversational interface itself.

  • NestJS
  • Node.js
  • Next.js
  • PostgreSQL
  • Elasticsearch
  • Redis

How we build it

  1. 01

    Study the workflow

    Look at how the work runs today and find where the time goes. Some jobs suit AI. Others are cheaper and safer as a few lines of ordinary code.

  2. 02

    Feasibility check

    A small trial on your real data. This answers the question that matters most: is AI accurate enough at this task? You get the answer before spending on a build.

  3. 03

    Prepare the data

    Gather the documents, clean them and split them so a model can search them. Answer quality depends on this step more than on the model.

  4. 04

    Build and measure

    Build the feature, then score it against a real set of questions. Without measurement there is no way to tell whether the next version is better or worse.

  5. 05

    Connect it up

    Wire it into the software, website or chat channel you already use. Permissions keep each person to the documents they are allowed to see.

  6. 06

    Hand over and monitor

    Source code and operating documentation handed over. We watch model spend and answer quality once it is running for real.

Projects we have built

F68

A stock analysis platform. It pulls together price data, financial statements and market news. From there it compares companies, assesses risk and builds investment reports in a conversational interface. Built on OpenAI and Anthropic models.

EzBook

A reading and listening app. Users scan a page with the camera or drop in a PDF. The system recognises the text in the image, turns it into readable text and reads it aloud. The hard part is that phone photos come in skewed, blurred or badly lit. Recognition quality rests largely on how the input image is handled first.

Frequently asked questions

Will AI replace our staff?
No, and that is not how Solv sells it. AI is good at repetitive work and at reading through a lot of documents. The final call should stay with a person, particularly where money and contracts are involved.
Does our data leave the company?
It depends on the setup. With the OpenAI or Anthropic APIs the content of a question does go out. We limit what is sent and document it at handover. Where data cannot leave your system at all, we run an open model on your own servers.
What if the chatbot answers wrongly?
Every model gets things wrong sometimes. We reduce the risk by making the chatbot answer only from your documents. Every answer cites its source, so a reader can check it. When something is not in the documents it says it does not know, rather than inventing an answer.
What does it cost to run?
There are two parts. The build itself, and the monthly model usage. The second is charged by volume of text processed, so it scales with how many people use it and how often. We estimate it during the feasibility check so you know up front, and set a cap so it cannot run away.
We are small with little data. Does it still work?
Yes. A knowledge chatbot is already useful with a few dozen documents. What decides the result is not the volume of data but whether the documents are written clearly.
How long does it take?
A feasibility check usually runs one to two weeks. An internal knowledge chatbot at a usable standard takes a few weeks to a month. Workflow agents take longer because they have to connect to systems you already run.
Do we get the source code?
Yes, as with every Solv project. You get the full source, the operating documentation and control of the model accounts. Changing development partners later is not a problem.

Wondering where AI would fit?

Tell us briefly which process eats the most time. We will come back with an honest read on whether AI can handle it, and how far.

Get in touch