AI tools for your team's work.
We help you decide which tasks to assign to language models. We review your processes, build a pilot, evaluate the results and train your team.
We use Cursor, Claude, GPT and Gemini in our daily work. In a consultation, we discuss your processes, explain AI's limitations and identify tasks where its usefulness can be tested.
How we can help
AI strategy
We review processes, select tasks for automation, assess potential returns and prepare a plan for the next 6–12 months.
RAG systems
Search systems for your company's documents. The language model uses retrieved material to prepare an answer and cite its sources.
AI agents
Agents for support, sales and internal operations. We connect tools, retain necessary information and set limits on what an agent can do.
Team training
Practical sessions with Claude, GPT and Cursor. Together, we prepare prompts and templates for your team's everyday tasks.
Fine-tuning
When an existing model falls short, we assess whether fine-tuning could help. We use SFT, DPO and distillation, evaluate quality and roll out changes in stages.
Evaluation & safety
We test responses against a set of queries and track errors and invented claims. We also test defenses against instructions introduced through external data.
Toolkit
We choose models and tools based on quality, cost, speed and data-hosting requirements. Our work includes both open models and hosted services.
How we run a project
Initial assessment
Over 1–2 weeks, we review processes, identify potential AI tasks and assess expected returns. You then decide whether to continue.
Pilot project
Over 1–2 months, we test a solution for one task, such as support. We launch it and measure the result.
Ongoing support
Engagements of three months or more to maintain and develop AI systems, train staff and work on new tasks.
Before we start
Where should LLM adoption start in a company?
Start with a short audit: 1–2 weeks to review your processes, map opportunities and estimate ROI. You're not committed to continue.
Which models do you use?
Claude Sonnet 4.6 and Opus 4.8, GPT-5.5, Gemini 3.1 Pro and Flash, and open models Llama 4, Qwen3, DeepSeek V4, Kimi K2.6 and MiniMax M3. We choose by quality, cost, latency and on-prem requirements.
How is RAG different from fine-tuning?
RAG retrieves relevant documents and passes them to the model with the question. This is useful when information changes frequently. Fine-tuning changes the model's parameters and may help it follow a particular style or response format. The two approaches can be combined.
How do you measure AI quality?
We collect real user queries and run checks before each release. We compare responses with the previous version, record errors and track invented claims. The results help us decide whether a change is ready to release.
What would you like AI to help with?
In a free one-hour consultation, we will discuss your processes and where to begin. We will explain how to assess the costs and potential benefits.