I design, evaluate and ship LLM features for teams that need them accurate, cheap to run and safe to expose.
Four projects, each with a number someone can check.
Voice AI phone line
Polish-language agent with a latency budget planned down to the millisecond.
AI dish videos for restaurant menus
Photo to five-second clip for about $0.51 per dish, vendors picked for commercial licensing.
Public tender analytics
537,000 tenders and 315,000 bids turned into win probability and pricing signals.
3D virtual tours at scale
81 apartment walkthroughs rendered on cloud GPUs, with cost tracked per scene.
Measure before shipping
A test set, a blind judge and a program that checks facts. If a change cannot be measured, it waits.
Most clients start with an assessment. It is the cheapest way to find out whether we should go further.
We go through your code, prompts and data together. You get a written opinion: what will break, what it will cost to run and what to measure first.
€850
Book a callOne AI feature taken to production with an evaluation suite, a cost model and monitoring your team can keep running.
from€9,000
I own the AI part of your product: roadmap, build, evaluation and the on-call for when a model changes under you.
from€12,000/ month
I'm Kacper Najda, an independent AI engineer. I build LLM features for companies and I run AI products of my own, so I pay for the tokens I recommend and live with the on-call that comes with them.
My work sits between product and infrastructure: retrieval, evaluation, cost control and the boring parts that keep a model honest in production.
Most likely. I write TypeScript and Python every day and work with models from OpenAI, Anthropic, Google and open-weight models. I fit into your repositories, CI and review process.
You do. Everything lives in your repositories from the first day.
I have built under GDPR article 9 for health data. Where it matters we use EU inference, providers that do not train on your data, and designs that keep sensitive fields out of prompts entirely.
Yes, for volume: exploring a codebase, writing tests, reviewing changes. Every change is still reviewed and tested before merge, and nothing reaches production without your approval.
Start with the Technical Assessment. It costs €850 and often answers whether you need a consultant at all.
Poland. I work remotely with teams across Europe, async-first, with overlap in EU business hours.
A 30-minute call, no pitch. If I am not the right person, I will tell you who is.
Tell me what you're buildingBook a callor write to [email protected]