Head of AI · ML Engineer · PhD Researcher

Stefan Seman

Production AI systems at the intersection of software and biomedicine.

I lead AI teams and still build the systems myself. My work spans LLM and RAG architectures, agentic applications, enterprise document intelligence, and cloud-native data platforms — grounded in a decade of biomedical research and peer-reviewed publication.

Portrait of Stefan Seman
Animated electrocardiogram trace
0+ Years in ML & data
0 Peer-reviewed works
0+ Citations
0 Systems delivered

Selected work

Systems, not demos.

Production AI and data platforms delivered for enterprise clients — each one taken from problem statement through to something running, monitored, and maintained.

Collaborations

Who I build with.

Delivery partners I build for through 404 Solutions, my B2B consultancy — agencies and product firms in Croatia, Bulgaria, and the US who bring me in for their enterprise clients.

Provectus Inc

Document Intelligence

United States

Document intelligence delivered together for a specialty insurance client — summarization and point-of-interest extraction, multi-document risk analysis, and the large-scale ingestion pipelines underneath both.

Generative AIAWSBedrock

WiserTech

ML Engineering

Bulgaria

Machine learning engineering delivered for a UAE-based sovereign AI organisation — speech-to-text pipelines, orchestrated agent graphs running over RAG retrieval nodes, and summarization that synthesizes insight across sources rather than returning them one at a time.

Speech-to-TextAgent OrchestrationRAG

Q Agency

Software Delivery

Croatia

Built and shipped a complete content platform — web search and deep-research agents feeding long-form generation, blog drafting, and publishing straight out to connected channels. Separately, mapped an undocumented production system and rebuilt its backend for one of their enterprise clients.

Agentic AIDeep ResearchContent Generation

Career

Ten years, one throughline.

From biomedical signal processing to leading an AI practice — the domain and the scope changed, the discipline of turning messy data into something dependable did not.

  1. Apr 2025 — Present Current

    Head of AI

    ExpertGroup

    • Lead the AI and Data team — organisation, people management, hiring and mentoring.
    • Own AI initiatives end to end: originating projects, presales, architecture, and engineering standards.
    • Stay hands-on — building demo pitches, proofs of concept and MVPs alongside the team.
    • Set technical direction and make delivery decisions across the AI portfolio.
    • Delivered RAG chatbots over live document collections — multilingual retrieval with answer lineage, and multilingual document QA.
    • Built gradient-boosted time series models for supply chain and sales forecasting.
    • Architecting crowd movement prediction and venue analytics for large-scale events.
    PythonFastAPILangChainQdrantRabbitMQAzure AI FoundryMicrosoft FabricAzure DevOpsDockerXGBoostComputer Vision
  2. Oct 2024 — Present Current

    Independent AI & ML Consultant

    404 Solutions

    • Run an independent practice delivering AI systems for partner organisations across several countries.
    • Designed and operated multi-tenant AI systems for enterprise clients.
    • Delivered document intelligence solutions for automated extraction, classification, and summarization.
    • Built scalable AWS workflows using S3, Step Functions, Bedrock, and APIs for multiple organizations.
    • Deployed, monitored, and maintained production systems across distributed client environments.
    PythonAI AgentsFastAPIRAGAWSBedrock
  3. Oct 2023 — Sept 2024

    Senior Data Scientist & Software Engineer in AI, R&D

    Publicis Sapient

    • Led development from proof of concept to production-grade systems.
    • Built enterprise RAG systems with retrieval optimization.
    • Delivered client-facing solutions and technical propositions.
    PythonFastAPILangChainTransformersPyTorchDockerAzureKubeFlowRedisMilvusAzure OpenAI
  4. May 2022 — Oct 2023

    Data Scientist & Developer → Senior Specialist

    Valcon Netherlands

    • Led and mentored a team of 6 engineers.
    • Owned design of document processing microservices.
    • Built data extraction and document classification services from scratch.
    PythonAzure DevOpsAzure Data FactorySynapseData LakeDocker
  5. Oct 2021 — Apr 2022

    Data Scientist

    Optimal Systems GmbH

    • Performed model evaluation and hyperparameter tuning for ML classification.
    • Produced data-driven insights and testing/validation for ML systems.
    PythonDockerMachine LearningAzure
  6. Mar 2020 — Mar 2022

    Researcher

    BioIRC — Bioengineering R&D Center

    • Built components for the SILICOFCM EU Horizon 2020 computational platform.
    • Worked on ECG, CPET, and biomedical data processing.
    • Contributed to peer-reviewed publications.
    PythonStatistical AnalysisSPSSECGCPETBiomedicine
  7. Oct 2018 — Mar 2022

    PhD Researcher

    University of Belgrade

    • Developed research methodology and statistical frameworks.
    • Applied machine learning to biomedical datasets.
    • Published papers internationally.
    PythonPyTorchSPSSMachine LearningBiostatistics
  8. Oct 2014 — Jul 2018

    Sports Scientist

    Sport and Recreation Center Obliqus

    • Built algorithms for signal processing and data collection.
    • Performed functional testing and performance analysis for athletes.
    PythonSPSSLabVIEWData Analysis

Research

Published, cited, applied.

Peer-reviewed work in computational cardiology and exercise physiology, with machine learning as the through-line.

  • 2026

    Reduced cardiac workload and enhanced oxygen extraction during maximal exercise in recreational endurance runners

    Open Heart · Journal Article

    Exercise Physiology
  • 2024

    Sacubitril/valsartan for the treatment of non-obstructive hypertrophic cardiomyopathy: an open-label randomized controlled trial (SILICOFCM)

    European Journal of Heart Failure · Journal Article

    Computational Cardiology
  • 2024

    The ventilatory efficiency parameters outperform peak oxygen consumption in monitoring the therapy effects in patients with hypertrophic cardiomyopathy

    Progress in Cardiovascular Diseases · Journal Article

    Computational Cardiology

Toolkit

What I build with.

The stack behind the systems above — weighted toward Python, cloud-native services, and the retrieval infrastructure that generative AI actually runs on.

Languages

  • Python
  • SQL
  • Bash
  • LabVIEW

ML & Generative AI

  • PyTorch
  • Transformers
  • LangChain
  • LlamaIndex
  • RAG
  • AI Agents
  • Fine-Tuning
  • Embeddings
  • XGBoost
  • Computer Vision

Backend & Services

  • FastAPI
  • Flask
  • REST APIs
  • Docker
  • PySpark
  • RabbitMQ

AWS

  • Lambda
  • S3
  • ECS
  • EC2
  • Step Functions
  • Bedrock
  • SAM

Azure

  • Data Factory
  • Synapse
  • DevOps
  • Functions
  • AI Search
  • Azure OpenAI
  • AI Foundry
  • Microsoft Fabric
  • Data Lake
  • Databricks

Data & Vector Stores

  • PostgreSQL
  • DynamoDB
  • Redis
  • Qdrant
  • Pinecone
  • Milvus
  • PgVector

Ops & Tooling

  • Git
  • Jenkins
  • Grafana
  • KubeFlow
  • Power BI

Research Methods

  • Biostatistics
  • SPSS
  • ECG Analysis
  • CPET Analysis
  • In Silico Trials

Education

  • PhD Candidate, Sport Science University of Belgrade · 2018 — Present Machine learning applications in exercise physiology and biomedical signal processing.
  • Master's Degree, Sports Medicine University of Novi Sad
  • Bachelor's Degree, Sport Science University of Belgrade

Certifications

  • Microsoft Certified: Azure AI Associate Microsoft
  • Microsoft Certified: Azure Data Fundamentals Microsoft
  • Lakehouse Fundamentals Databricks Academy
  • Generative AI Fundamentals Databricks Academy
  • COVID-19 Contact Tracing Johns Hopkins University (Coursera, 2021)
Animated electrocardiogram trace

Get in touch

Building something that needs to actually ship?

I work on production AI systems — LLM and RAG architectures, document intelligence, and the cloud plumbing that keeps them running. Open to conversations about engineering work and research collaboration alike.