AI ENGI // applied NEERING

Computer vision, deep learning, and multimodal AI systems — engineered for production.

SERVICES

// what we do

01

Generative & Multimodal AI

LLMs, vision-language systems, diffusion pipelines, and custom generative workflows designed around real product and business needs.

// When off-the-shelf generative models can’t meet brand, domain, or compliance constraints.

LLM, diffusion, or VLM pipeline
Domain-tuned models & evaluation
Production-ready API or integration

Available

02

Deep Learning Systems

Custom architectures, training pipelines, optimisation, quantisation, and deployment at scale.

// When off-the-shelf models hit accuracy, latency, or cost ceilings.

Custom architecture & training pipeline
Inference optimisation & quantisation
Deployment, monitoring & iteration

Available

03

Computer Vision

Object detection, segmentation, tracking, 3D reconstruction, biometric verification. From dataset to deployed model.

// When you have visual data — images, video, scans — and need reliable automated understanding.

Dataset curation & annotation strategy
Model training, evaluation & iteration
Edge or cloud inference, integrated

Available

04

Research → Production

End-to-end AI engineering: problem framing, research, implementation, and production deployment with rigorous evaluation.

// When the problem is hard enough that ready-made tools won’t solve it.

Problem framing & literature review
Prototyping with rigorous evaluation
Production implementation & handover

Available

// capabilities Computer Vision Multimodal AI Deep Learning Systems Research to Production

// tooling PyTorch OpenCV CUDA ONNX FastAPI Docker

PROJECTS

// selected work

P.01 2026 · ongoing

Real-time 3D Avatar System

Computer Vision · 3D · NLP · Audio

Research & Implementation

Client: Confidential

Mainstream platforms leave deaf users with flat captions — text that strips the grammar, timing, and non-manual markers that make sign language readable. Nexome is building a real-time 3D avatar driven by an LLM for semantic translation, paired with a motion-synthesis pipeline that produces natural, expressive signing from text or audio. Ongoing engagement on the client’s platform.

Motion synthesis pipeline from linguistic input Custom skeletal animation driven by ML model Integrated with client's production platform

P.02 2024-25

Biometric Verification Pipeline

Computer Vision · Biometrics · Multimodal LLM

Full-stack AI Engineering

Client: Confidential

KYC-regulated industries need identity verification that holds up against motivated attackers, not just casual fraud. Co-built by Nexome’s founder: a multi-layer pipeline (face verification, document OCR and authentication) engineered to catch presentation attacks (screen and print) and document modifications across text, photos and security features. In production, sub-200ms inference on standard hardware, GDPR-compliant by design.

Custom deep learning architecture for face and document verification Sub-200ms inference on standard hardware GDPR-compliant data handling & audit compliance

P.03 2025

Medical Imaging Synthetic Data Pipeline

Computer Vision · Generative AI · Medical Imaging

Research & Implementation

Client: Confidential

Eye disease detection models stall on rare cases — the very ones where AI matters most — because clinical datasets are small and skewed. Nexome combined realistic medical image synthesis (SDXL with LoRA fine-tuned on limited clinical data) with classification training, producing balanced datasets that cover rare presentations without violating data-scarcity or compliance constraints.

Realistic image generation via SDXL + LoRA trained on limited clinical data Anatomically consistent disease simulation using segmentation-guided inpainting Balanced dataset construction with metadata & classification model training

ABOUT

// who we are

// founder

AI ENGINEERING. EXPERTLY.

Nexome is a founder-led AI engineering practice run by Michał Budnik, built around deep expertise in computer vision and applied deep learning.

The work spans the full stack: research, model design, training, optimisation, and production deployment.

LinkedIn ↗ mbudnik@nexome.eu

Work with Nexome →

// profile

Founder Michał Budnik

Expertise Computer Vision · Deep Learning · Multimodal AI

Approach Research quality · Production ready

Compliance EU · GDPR-aligned

Location Poland · EU · Remote

CONTACT

// let's talk

LET'S BUILD SOMETHING.

Describe the challenge. Response within one business day.

✓

MESSAGE SENT.

We'll be in touch within one business day.