Research

Protein AI & Mechanistic Interpretability

Conditional Protein Sequence Generation

Extending diffusion models on protein language model encodings for controllable biological sequence generation.

Mechanistic Interpretability

Understanding what protein language models learn using Sparse Autoencoders (SAEs) and protein concept-based probing.

Multimodal Controllability

Combining sequence, structure, and free-text functional annotation modalities for enhanced generative control.

Key Technologies

ESM-2, ESMFold, Diffusion Models, Sparse Autoencoders, Concept Bottleneck Models, GFlowNets, Circuit Discovery.

Pieris Kalligeros

Pieris Kalligeros

PhD Research Candidate

Email GitHub LinkedIn Scholar Blog Art

Building at the intersection of generative modelling and mechanistic interpretability — applied to proteins. My research spans computer science, robotics, and biosystems engineering, converging on designing controllable, interpretable generative models for biological sequences.

CV

Education, work, competitions & seminars

pierisk98@gmail.com
github.com/pieris98
linkedin.com/in/p-kalligeros

Education

  • MSc Artificial Intelligence — KAIST (2022–2025). Thesis on GFlowNets for compositional generation of 3D geometric objects and energy manifolds.
  • MSc Biosystems Engineering — Wageningen University & Research (2021–2022, on hold). 60 ECTS complete.
  • BEng Computer Science — University of Edinburgh (2016–2020). First Class Honours.

Work Experience

  • PhD Research Candidate — Archimedes Research Unit, Athena RC / AUEB (2024–Present). Protein AI and mechanistic interpretability.
  • AI & Computer Vision Engineer — Zentech (2020–2025). Real-time AI pipelines for motorway enforcement and video tolling.
  • Research Intern — Vision-Robotics, CyRIC / PetaParsec (2024). Vision-robotics grasping of texture-less objects.
  • Research Intern — LFCS, University of Edinburgh (2019). HCI research; co-authored IEEE EuroS&P paper.

Selected Competitions

  • OpenHackathons N-Ways GPU CUDA Bootcamp — EuroCC2 & NVIDIA (2024).
  • QuHackEd 2019 — Quantum ML recommender system using Rigetti Computing.
  • HackUPC 2019 — ClimAid, a climate-impact awareness mobile app.
  • HackMoscow 3.0 — Deklarator API visualisation web app.

Conferences & Seminars

  • Greeks in AI (2025–Ongoing), Athens.
  • European Bitcoin Convention 2024, Barcelona.
  • NVIDIA GTC (2020–Ongoing), online.
  • Panhellenic Conference on Informatics, PCI 2020.
  • VivaTech, Paris (2018).