Skip to content
View lighteternal's full-sized avatar

Block or report lighteternal

Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
lighteternal/README.md
Header banner

Hi there, I'm Dimitris!

Typing SVG

GitHub followers GitHub User's stars Profile Views


About Me

Senior AI leader and creator of llmarena.gr and drugbase.gr.

An engineer at heart with 10+ years building AI systems, leading AI strategy and teams, and turning research into systems people can trust. I stay close to the code to make sure what we build is fair, transparent and useful.

My journey has taken me through cybersecurity, biomedicine and fintech: three very different arenas, united by one idea: when AI is built for people, it transforms entire industries. I've led and built autonomous systems that safeguard digital assets, surface breakthrough biomedical insights and restore fairness in debt collection. Technology is the tool; empowerment is the goal.

As a mentor and storyteller, I believe in the power of continuous learning and collaboration. My roots in research and machine learning (a PhD in Machine Learning focused on NLP) shaped how I solve problems, and I love sharing that journey with others.

πŸ“ Location     Athens, Greece
πŸŽ“ Background   PhD in Machine Learning (NLP)
🧠 Interests    Multi-agent AI, biotech, fintech, NLP, knowledge graphs
πŸ€— Handle       lighteternal on Hugging Face

Focus Areas

Python Agentic AI Multi-Agent Systems Biomedical AI Fintech AI NLP Research Neo4j Hugging Face RAG Knowledge Graphs Predictive ML


Featured Work

Project What it explores
BioDecision-4B System-1 decision model for biomedicine, pharma and clinical trials: returns a calibrated probability per answer in one forward pass, with new decision tasks defined in the request. 70.2% on 46,199 held-out decisions from 28 biomedical benchmarks (+7.8 over Qwen3.5-4B). Demo Β· Data.
llmarena.gr Greek LLM Benchmark β€” head-to-head evaluation and leaderboard of large language models on Greek language tasks. Live at llmarena.gr.
drugbase-gr Open corpus of every authorized pharmaceutical in Greece: 5,018 brands, 13,445 structured documents from Ξ•ΞŸΞ¦ and EMA. Live at drugbase.gr.
legalize-gr Version-controlled Greek legislation corpus β€” 6,804 Ξ¦Ξ•Ξš Α΄ laws and 26 consolidated codes as Markdown, git-tracked by publication date.
PsychGNN Graph neural network for psychiatric genomics β€” link prediction and masked edge imputation over gene-phenotype networks. On Hugging Face.
BioAssayAlign Drug-discovery embedding model for bioassay ranking and assay-conditioned molecular representation, built on Qwen3. On Hugging Face.
dendrite An open-source, agentic biomedical discovery workbench for evidence-attributed research synthesis.
jury Multi-agent and decision-intelligence experimentation around agent orchestration and evaluation.
FarFetched_NLP Claim validation in dynamic textual environments with an entity-centric reasoning approach.
PENELOPIE Greek-language open information extraction through machine translation, grounded in NLP research.
Automated-Knowledge-Graph-Construction Entity extraction and automated knowledge graph construction with NLP and Neo4j.

Research Notes

  • PhD in Machine Learning with a focus on Natural Language Processing
  • Research and applied work across biomedical AI, knowledge graphs, claim validation, and Greek-language NLP
  • Publications and research artifacts available across Google Scholar and Hugging Face

Snake animation


Connect With Me

GitHub LinkedIn Hugging Face


Current Focus

πŸ”­ Building      Agentic AI for regulated decision, credit risk, and research workflows
🧩 Exploring     Multi-agent systems, deep-research agents, biomedical copilots, and evaluation loops
🌍 Interested in Practical GenAI and predictive ML that augment experts without adding operational risk

Random Quote

Readme Quotes


Footer banner

From NLP research to agentic AI in biotech and fintech.

Pinned Loading

  1. drugbase-gr drugbase-gr Public

    Structured corpus of all authorized pharmaceutical products in Greece (Ξ•ΞŸΞ¦ + EMA). 5,018 brands, 13,445 documents, CC BY 4.0.

    11

  2. dendrite dendrite Public

    TypeScript 1

  3. PENELOPIE PENELOPIE Public

    PENELOPIE: Enabling Open Information Extraction for the Greek Language through Machine Translation

    Jupyter Notebook 7 1

  4. legalize-gr legalize-gr Public

    Greek legislation in Markdown β€” Ξ¦Ξ•Ξš Α΄ 2000–2026 (6,804 laws) + consolidated codes (26 codes, 6,359 articles)

    50 5

  5. FarFetched_NLP FarFetched_NLP Public

    A claim validation framework for dynamic textual environments

    Jupyter Notebook 2

  6. CORD-19-OIE-triple-extraction CORD-19-OIE-triple-extraction Public

    A dataset containing CORD-19 related OIE triples from our information extraction pipeline, linked to UMLS entities.

    2