From infrastructure
to intelligence.
Applied AI builds the full stack — the cluster, the services, the models — so AI actually survives contact with production.
// what we build
Three layers,
one team
Most AI projects fail not because of the model — but because of everything around it. We own the full stack from the platform up, so nothing falls between the cracks.
One team, zero hand-offs. The people who train the model also run the cluster it lives on.
AI Layer
Custom models — training, fine-tuning, and deploying ML/NLP models for problems where off-the-shelf doesn't cut it. Applied AI — using existing AI tools to create new business products. Not research for research's sake — we build things that ship.
Backend Services
Production backend systems in Go — APIs, data pipelines, ETL, real-time processing. The plumbing that connects AI to the rest of the business, and the web front ends on top of it.
Infrastructure
Cloud-native infrastructure on Kubernetes. Deployment, scaling, networking, and operations for AI workloads. We run the platform so the models have somewhere to live.
// track record
Shipped, not
hypothetical
CityChat
Conversational AI city guide — our own product, live on iOS & Android in 16 cities across Poland and Spain. Native apps, a Go backend with geospatial and vector search, and LLM chat grounded in a curated database of 35,000+ places — built and operated end to end. citychat.pl ↗
2026 — now
FormFlow
Multi-brand contact-form platform with a three-stage AI image-authenticity pipeline, built end to end for a leading Polish food & beverage group. formflow.solutions ↗
2026 — now
SanAlert
Unofficial early warning about air threats over Poland — drones and missiles tracked across the border, a threat level for the place you choose, and official RCB and RSO messages in their original wording. A finished product: web app and native iOS, built and run end to end. sanalert.pl ↗
2026 — now
JPK Doctor
A validator that answers the only question an accountant actually has: will the tax authority's gateway accept this SAF-T file? Schema checks plus the undocumented gateway rules, returned as one verdict — accepted, likely to trigger an automated query, or rejected — instead of a wall of errors. Written in Rust, runs offline: the file never leaves the disk. jpk-doctor.pl ↗
Wasabi AiR
Built a scalable AI-powered media search engine from scratch — Go backend, ML-driven indexing and retrieval. The archive is indexed second by second: faces, logos, objects, on-screen text, sound events and music, and content moderation for nudity, violence and weapons — plus speech-to-text with speaker diarization in 180+ languages. Every match comes back as a timecode, so a search lands on the frame, not the file. Production system serving enterprise customers.
GrayMeta
Deployed AI platforms in secure enterprise environments. Integration of ML models into existing corporate infrastructure.
AT&T / DirecTV
ETL systems processing hundreds of thousands of events per second. Large-scale data pipeline engineering.
// product
Built for one,
ready for many
The platform we built for that client is now a product. Contact forms for consumer brands with a three-stage AI image-authenticity pipeline: brand teams manage forms without engineers, and quality teams get every submitted photo verified — authentic, suspect, or doctored — before anyone opens the ticket.
// the team
Two engineers,
full ownership
// ceo — business & product
Paweł Rojek
AI Engineer
Software engineer and leader with hands-on experience delivering complex tech projects. Drives business strategy, product development, and operations. Deep understanding of AI/ML combined with experience in commercial deployments.
pawelrojek.dev ↗// cto — architecture & engineering
Piotr Rojek
Principal Engineer
Software architect with 15+ years of experience. Owns technical strategy, system architecture, and engineering leadership. Specializes in large-scale systems, NLP, and applying AI in commercial products. Core expertise in Go and cloud-native architectures.
piotrrojek.io ↗// how we work
Three ways
to engage
Discovery sprint
A short, fixed-scope engagement to map the problem, prove feasibility on your data, and leave you with a concrete plan — whether or not you build it with us.
End-to-end delivery
We own the whole thing — infrastructure, backend, models — from first commit to production. One team accountable for the outcome, not a layer of it.
Team extension
We embed with your engineers to accelerate an AI effort that's already moving — architecture, platform work, or model engineering where you need depth.
// get in touch
Have a problem that
needs shipping?
[email protected]
We read everything. Serious problems get a reply from an engineer, not a funnel.
