Open Thesis/Internship Positions: Join AI4IA 

About AI4IA

AI4IA is a cross-disciplinary initiative focused on the scientific study of AI adoption dynamics, digital innovation ecosystem governance, organisational transformation methodologies, and the design of evidence-based policy instruments.

Our research agenda investigates the mechanisms by which AI capabilities translate — or fail to translate — into measurable organisational and societal value. This includes the design and validation of AI transformation methodologies, the modelling of digital maturity trajectories across organisations, the governance of federated AI ecosystems such as AI Factories and data spaces, and the formulation of evidence-based policy frameworks for national and European AI strategy. The group does not limit itself to the study of these phenomena but actively develops the infrastructures and tools that operationalise them: from sovereign data ecosystems such as the Demokritos Dataspace and PHAROS Data Lab — which enable trustworthy data sharing and experimentation— to intelligent innovation support tools.

AI4IA positions itself as a “full-stack actor in the AI adoption chain”. The group’s research outputs, methodological frameworks, and collaborative structures are designed to generate institutional capacity that persists beyond the duration of any individual project or funding cycle.

Open Thesis/Internship Positions – Overview

We are currently looking for motivated undergraduate or postgraduate students to join our team in Agia Paraskevi, Attica, Greece (Hybrid). Below is the list of available thesis/ internship topics:

• Topic 1: Explainable AI (XAI) for Business Adoption

• Topic  2: Federated Data Space Architecture 

• Topic  3: DCAT-AP Metadata Catalogue for the Demokritos/PHAROS Data Lab

• Topic  4: Data Lab Curation Pipelines for AI-Ready Datasets

• Topic 5: Scalable Data Ingestion for Environmental and Heterogeneous Sources

• Topic 6: Quality Assessment Framework for the Demokritos/PHAROS Data Lab

Detailed Topic Descriptions

Topic 1: Explainable AI (XAI) for Business Adoption

The deployment of AI systems in organisational settings increasingly depends not only on technical performance but on the degree to which decision-makers can interpret, trust, and act upon model outputs. Despite rapid advances in explainability methods, the relationship between explanation quality and actual adoption remains poorly theorised and empirically underexplored. A technically sound explanation does not automatically translate into organisational uptake; factors such as user role, decision context, sector-specific accountability requirements, and institutional readiness all mediate whether XAI mechanisms function as genuine adoption enablers.

This thesis investigates XAI as a structural bridge between algorithmic outputs and the organisational conditions that govern AI integration. Rather than optimising explainability for its own sake, the work is anchored in adoption science: it asks which explanation types reduce which adoption barriers, for which users, and under what contextual conditions. The candidate will examine how feature attribution, causal explanation, and contrastive reasoning techniques map onto constructs such as perceived transparency, algorithmic trust, and delegation of decision authority — drawing on sectors where these tensions are acutest, including finance, healthcare, and public administration.

Tasks. The candidate will contribute to one or more of the following threads: (i) empirical mapping of XAI technique types to adoption barrier profiles across user roles and deployment contexts; (ii) design and validation of instruments measuring explanation quality from an end-user rather than a purely technical perspective; (iii) mixed-methods analysis — survey, interview, or configurational approaches such as fsQCA — of cases where XAI is a significant mediating factor in AI adoption outcomes; (iv) contribution to a causal framework formalising the relationship between XAI affordances and adoption constructs. The work is embedded in the AI4IA group's active research programme and connects to EU-level deployments through the Pharos AI Factory and the European AI-on-Demand Platform.

Indicative Bibliography:

• https://link.springer.com/article/10.1007/s10994-021-05981-0

• https://www.sciencedirect.com/science/article/pii/S0040162522006412

• https://www.science.org/doi/10.1126/science.abg1834 https://aisel.aisnet.org/ecis2024/track03_ai/track03_ai/9/

Topic 2: Federated Data Space Architecture  

Objective. Define a reference architecture for the Demokritos Data Space, specifying its core layers and how Data Labs participate as data providers.

Description. The Demokritos Data Space is being designed as a governance and federation framework that enables multiple organizations to share datasets under controlled terms. Within this framework, individual Data Labs (such as PHAROS) act as the operational layer that prepares and exposes curated datasets. This thesis produces the architectural blueprint that all other Data Space components reference. It will adopt IDS-RAM 4 and Gaia-X Trust Framework conventions, align with the EU Data Spaces Support Centre blueprint, and clearly position PHAROS as the data front-end of the Greek EuroHPC AI Factory while preserving the separation of concerns between Data Lab (curation/preparation) and AI Factory (orchestration).

Tasks • Review IDS-RAM 4, Gaia-X Trust Framework, and EU Data Spaces Support Centre blueprints • Define the layer model (connector, catalogue, identity, governance, federation) • Specify integration points with the AI Factory orchestration layer, with no overlap of responsibilities • Define how a participating Data Lab exposes its DCAT-AP catalogue to the Data Space • Deliver architecture diagrams, deployment model, and prototype scaffolding (docker-compose)

Expected Outcome • Reference architecture document • Component-level deployment diagrams (three-tier vertical layout) • Scaffolding repository with docker-compose for core services

Required Skills • Distributed systems • Python • Technical writing

Recommended Technologies • Docker • Kubernetes • FastAPI • Eclipse Dataspace Components

Topic 3: DCAT-AP Metadata Catalogue for the Demokritos/PHAROS Data Lab

Objective. Build a searchable metadata catalogue for the Demokritos/PHAROS Data Lab, fully compliant with DCAT-AP 3. 0. Description. DCAT-AP is the European application profile of W3C DCAT and is the de facto interoperability requirement for data catalogues that participate in EU Data Spaces. This thesis delivers the catalogue that exposes the curated datasets of the Demokritos/PHAROS Data Lab and makes them discoverable within the Demokritos Data Space. The catalogue is the contact surface between Data Lab consumers (researchers, AI Factory workloads) and the underlying curated data. The work explicitly excludes orchestration and access enforcement, which belong to other components.

Tasks • Implement the DCAT-AP 3.0 metadata model (dataset, distribution, data service) • Build an ingestion pipeline for metadata records produced by the curation pipelines • Implement search and faceted browsing over the catalogue • Expose RDF and JSON-LD endpoints conformant to DCAT-AP serialisations • Validate the catalogue against the official DCAT-AP SHACL shapes

Expected Outcome • DCAT-AP 3.0 compliant catalogue service • SHACL validation report • Search interface with faceted browsing

Required Skills • Python • Semantic data models • REST APIs

Recommended Technologies • CKAN or Piveau • Python • ElasticSearch • PostgreSQL

Topic 4: Data Lab Curation Pipelines for AI-Ready Datasets

Objective. Build the curation and preparation pipelines that transform raw data into AI-ready datasets ready for publication in the Data Lab catalogue.

Description. The Demokritos/PHAROS Data Lab is the curation and preparation layer of the broader infrastructure. It is not the orchestration middleware: orchestration is the responsibility of the AI Factory. This thesis implements the dataset curation workflow: cleaning, schema harmonisation, enrichment with metadata, validation against quality criteria, and registration in the DCAT-AP catalogue. The pipelines must be reproducible and version-controlled. Human-in-the-loop steps must be explicit, particularly where GDPR constraints prevent fully automated decisions.

Tasks • Define curation workflow stages with explicit handoffs • Implement cleaning, harmonisation, and enrichment modules • Integrate with the quality assessment framework • Implement metadata registration in DCAT-AP • Document the boundary between automated processing and human review

Expected Outcome • Curation pipeline framework • At least one end-to-end pipeline applied to a real dataset • Documentation distinguishing automated steps from reviewed steps

Required Skills • Python • Data engineering • Data pipelines

Recommended Technologies • Apache Spark or Airflow • Pandas • Great Expectations • Python

Topic 5: Scalable Data Ingestion for Environmental and Heterogeneous Sources

Objective. Build ingestion pipelines that bring batch, file-based, and streaming data into the Demokritos/PHAROS Data Lab.

Description. The Data Lab must ingest data from heterogeneous sources before curation can occur. This thesis implements ingestion for a representative mix: Copernicus CAMS atmospheric composition products (batch and near-real-time), Greek national environmental monitoring data (file-based, heterogeneous formats), and a streaming source such as MQTT or Kafka for sensor data. The work focuses on robust, observable ingestion with schema validation and lineage tracking.

Tasks • Implement Copernicus CAMS ingestion via the CDS API • Implement file-based ingestion for national environmental monitoring datasets • Implement a streaming ingestion path (MQTT or Kafka) • Add schema validation, lineage capture, and observability metrics • Benchmark ingestion throughput under representative load

Expected Outcome • Operational ingestion service with three working source connectors • Lineage and metrics dashboard • Throughput benchmark report

Required Skills • Python • ETL pipelines • Networking basics

Recommended Technologies • Apache Airflow • Kafka • MQTT • Python • MinIO Mode: Standalone. Produces a working ingestion service usable independently.

Topic 6: Quality Assessment Framework for the Demokritos/PHAROS Data Lab

Objective. Build a framework that assesses dataset quality and surfaces results to curators, supporting human decisions on dataset acceptance.

Description. In a GDPR-constrained environment, dataset quality decisions cannot be fully automated, particularly when those decisions affect downstream AI training. This thesis builds a quality assessment framework that runs deterministic and statistical checks, including missing values, schema conformance, outliers, drift, and fitness-for-purpose, and produces assessment reports for human review. The framework defines quality gates as decision points reviewed by curators, not as automated remediation steps.

Tasks • Define a quality dimension model (completeness, validity, consistency, timeliness, fitness) • Implement statistical and rule-based checks for each dimension • Generate per-dataset assessment reports • Integrate with curation pipelines as advisory gates • Document the human-review workflow

Expected Outcome • Quality assessment toolkit • Standard report template • Documented human-review workflow

Required Skills • Python • Statistics • Data engineering

Recommended Technologies • Great Expectations • Pandas • Python

How to Apply (General Instructions)

To apply for any of the positions above, send your application to innovation [at] iit.demokritos.gr.

Your application must include:

1. Your CV: Please include links to your GitHub profile, open-source contributions, or any technical projects you have worked on, if applicable. 

2. A short Cover Letter: Maximum 1 page, explaining why you are interested in the specific topic and how it aligns with your skills and goals.

The file names should contain your last name, e.g. CV_Papadopoulou.pdf

Important - Email Subject Line:

Please refer to the topic you are applying for in the subject line and specify if you are interested in a thesis (jointly supervised with your university) or internship, e.g. “Application for Master Thesis - Topic X”