Skip to content

I build interpretable ML systems that ship

AI & Data Science consultant; PhD candidate at Aalborg University. 7+ years in predictive modeling, healthcare AI, time-series, recommenders, and signal processing. Published in CSBJ, CVIU, ICIP, and PLOS ONE. Kaggle medalist; reviewer for SIGMOD, ICDE, and VLDB.

  • Time-series (EHR, IoT, tabular) • Uncertainty • Model monitoring
  • MLOps-ready code, reports, and handoff your team can own
  • Built for outsourcing PMs & hiring managers: clear scope, timelines, and updates
Book a consult Download CV

Trusted by / Featured in

CSBJ (2025)
PLOS ONE
CVIU
Kaggle Top 7% (154/2435)
Reviewer: SIGMOD/ICDE/VLDB

Services & outcomes

Predictive modeling

Time-series (irregular sampling), tabular, multimodal. Emphasis on interpretability & uncertainty.

  • Win: higher AUROC/PR, fewer false alarms
  • Artifacts: codebase, experiment report, model card

Clinical & regulated AI

EHR/ICU modeling; clinician-in-the-loop; privacy-first.

  • Win: decision support clinicians can trust
  • Artifacts: validation plan, bias & drift checks

Data platform & handoff

Schema design, ETL, experiment tracking, monitoring hooks.

  • Win: reproducible workflows, faster iteration
  • Artifacts: docs, dashboards, onboarding guide

Highlighted work

Healthcare • Time-series

Glucose prediction for ICU patients

Irregular EHR modeling with integrated representation; built proof-of-concept demo for clinicians.

  • Evidence: CSBJ 2025; ScienceDirect; live demo available
  • Role: Lead modeling & evaluation
arXiv Code Demo
Vision • Retrieval

Graph hashing for image search

Reproducible spectral hashing implementations and a non-alternating graph hashing algorithm.

  • Evidence: Peer-reviewed: CVIU 2022, ICIP 2021
ESH code Hashing suite
Startup • Recommenders

Movie/TV recommendation platform

Co-founded startup; designed PostgreSQL schema; built Python/Django ETL with logging & caching.

  • Outcome: Delivered functional MVP ready for iteration
BCI • Signal processing

Knowledge translation for SSVEP BCIs

Used cross-subject information to improve target detection; increased ITR.

  • Outcome: ~+10% ITR over best baseline
Paper

Process & engagement

1) Discovery Sprint

1–2 weeks to clarify use-case, data, and success metrics.

Deliverables: roadmap, baselines, feasibility memo.

2) Feasibility Study

3–6 weeks to prototype, run ablations, quantify impact.

Deliverables: code repo, experiment report, model card.

3) Pilot / Demo

Interactive app or API; monitoring hooks.

Deliverables: demo, dashboard, handoff guide.

4) Productionization

Harden the solution: tests, CI, drift checks, observability.

Deliverables: deployable service, runbook, metrics plan.

Pricing: contact for quote.

Selected publications

  1. Enhancing glucose level prediction of ICU patients through irregular time-series analysis and integrated representation. CSBJ, 2025. ScienceDirect · arXiv
  2. Subject-independent information for SSVEP-based BCIs. PLOS ONE, 2020. DOI
  3. A non-alternating graph hashing algorithm for large-scale image search. CVIU, 2022.
  4. A simple supervised hashing algorithm using projected gradient and oppositional weights. ICIP, 2021.

More on Google Scholar.

Awards & achievements

  • Kaggle Bronze — Top 7% (154/2435, Severstal Steel Defect Detection), 2019
  • 3rd prize — National BCI Competition (70 teams), 2018
  • 3rd prize — Stock price prediction challenge (48 teams), 2017
  • Top 1% — National MSc entrance exam (Biomedical), 2015

FAQs

Do you sign NDAs? How do you handle data privacy?

Yes. I follow least-privilege access, encrypted storage, and can work with synthetic data when needed. Compliance: GDPR only; I align with EU data-processing requirements and clinical governance where applicable.

What’s your typical stack?

Python (PyTorch, scikit-learn), time-series libs, Postgres, Pandas/Polars; Weights & Biases; Streamlit/FastAPI for demos; Docker.

How do we start?

Fill out the contact form with your use-case, data shape, and success metric. I’ll reach out within 1–2 business days with next steps and (if helpful) a short written proposal.

Availability?

Limited client slots while finishing PhD; part-time retainers or project blocks work best.

Let’s build AI solutions together

I reply within 1–2 business days. GDPR-compliant workflows by default.