Vladimir Salnikov

Education

• Technical University of Denmark — MSc — Human-Centered AI, Visual Computing

  Aug 2025 – Jun 2027

  Lyngby, Denmark

• Lund University — BSc — Mathematics

  Sep 2022 – Jun 2025

  Lund, Sweden

Experience

• Student Researcher — Aarhus University

  Mar 2026 – Present

  Aarhus, Denmark

  Post-training of Danish foundation models.

  • Implemented and improved RLAIF algorithms for low-resource languages.
  • Built pipelines to generate synthetic instruction-tuning datasets for Danish.

• Computer Vision Intern — Kravi Analytics

  Mar 2025 – Present

  Lund, Sweden

  Computer vision for ocean surveillance and maritime technology.

  • Fine-tuned a foundation model to F1 0.92 on a false-positive estimation task for ship detections.
  • Used LoRA to fine-tune a Stable Diffusion model and built an automated synthetic data generation pipeline.

• Research Intern — CVML group, Lund University

  Jul 2024 – May 2025

  Lund, Sweden

  Relative depth estimation from dense feature matches under Viktor Larsson — the foundation of my Bachelor's thesis.

  • Trained Graph Neural Networks and CNNs for relative depth estimation (PyTorch, PyTorch Geometric, OpenCV).
  • Compared the approach against existing methods together with PhD students in the group.
  • Surveyed state-of-the-art literature on relative depth estimation and feature matching.

• AI Engineer Intern — Hello Ada AI

  Oct 2024 – Feb 2025

  Copenhagen, Remote

  AI-assisted programming platform for kids in Danish schools.

  • Evaluated speech-to-text models for Danish.
  • Implemented a text-to-code feature using reasoning LLMs, prompt engineering, and RAG.
  • Managed and deployed LLM models on Google Cloud (Compute Engine, Vertex AI).

Open-source projects

• Scandinavian Instruction-Tuning Datasets

  Jan 2026

  Python · Hugging Face Datasets · pandas · nb-nordic-lid

  Unified-format instruction-tuning datasets for Danish, Swedish, Norwegian Bokmål, and Nynorsk. Preprocessing pipeline for language ID, filtering, and chat-schema normalization.

• SMK Image-Text

  Dec 2025

  Python · Hugging Face Datasets · pandas · SMK API

  Image–text dataset from digitized SMK (Statens Museum for Kunst) artworks with Danish and English descriptions.

• Danish Wikipedia — Image, Caption, Context

  Sep 2025

  Python · HF Datasets · pandas · Wikipedia API

  First large-scale Danish image–text dataset that keeps neighboring article context, not just a single caption.

Skills

• Languages Python · R · Matlab · C++ · Haskell · Fortran
• ML & data PyTorch · PyTorch Geometric · NumPy · pandas · scikit-learn · OpenCV · Matplotlib · Hugging Face Datasets
• Cloud Google Cloud (Compute Engine, Vertex AI) · Microsoft Azure · Hugging Face Hub
• Spoken Russian (native) · Ukrainian (native) · English (C1) · Swedish (A2) · German (A2) · Danish (A1)