Panos Kolyvakis

Experience & Education

From the lab to agentic AI.

Eight years of research and teaching at the University of Edinburgh, then data science in R&D and FinTech, and now leading agentic AI and policy-as-code governance at Kyndryl.

View my CV (PDF)

Work experience

  1. Jul 2026 – present

    Associate Director, Global AI and Data Science

    Kyndryl · London, United Kingdom

    Leading the delivery and client integration of Kyndryl's Policy as Code capability. Developing a policy engine that extracts rules from business documents, converts them into Rego, and enforces customer-specific guardrails through Open Policy Agent (OPA). Integrating the capability into customer environments with auditable policy decisions and deployments using Terraform and Argo-based CI/CD.

    • Policy as Code
    • Open Policy Agent
    • Rego
    • Terraform
    • Argo CD
    • Agentic AI governance
  2. Dec 2025 – Jul 2026

    Senior Lead, Global AI and Data Science

    Kyndryl · London, United Kingdom

    Built an enterprise agentic framework from scratch in pure Python, including function calling, an orchestration engine, dynamic tool discovery and execution, agent memory, concurrency, and configuration-driven MCP servers, registry and gateway. Delivered end-to-end observability using OpenTelemetry (OTel), OpenLLMetry, Prometheus, Grafana Mimir and dashboards.

    • Python
    • Agentic AI
    • MCP
    • Orchestration
    • OpenTelemetry
    • OpenLLMetry
    • Prometheus
    • Grafana Mimir
  3. Sep 2024 – Nov 2025

    Senior Data Scientist, AI / ML

    Traydstream · London, United Kingdom

    Trained and productionised ML models for FinTech, focusing on both text and multimodal LLMs.

    • Fine-tuned multimodal LLMs with full-parameter fine-tuning, LoRA and QLoRA.
    • Optimised object-detection pipelines (CNNs) on Azure VMs and built efficient data pipelines.
    • Developed APIs with FastAPI and Flask, and managed Docker environments.
    • Worked on agentic AI for financial compliance checks, rigorously evaluated LLM outputs, and ran VM-hosted Transformer models with Hugging Face and vLLM.
    • Multimodal LLMs
    • LoRA / QLoRA
    • CNNs
    • Azure
    • FastAPI
    • Docker
    • Hugging Face
    • vLLM
  4. Feb 2024 – Sep 2024

    Data Scientist, Research & Development

    Twyn Holdings Limited · London, United Kingdom

    My work focused on LLMs: prompt engineering techniques such as few-shot and chain-of-thought for multi-label classification, fine-tuning LLMs, and LangChain for multi-agent systems and RAG pipelines. I fine-tuned and maintained models in production on Azure OpenAI and Azure Machine Learning Studio.

    • LLMs
    • Prompt engineering
    • Fine-tuning
    • LangChain
    • RAG
    • Azure OpenAI
    • Azure ML Studio
  5. Mar 2023 – Nov 2023

    Business Intelligence / Data Analyst

    Xenos Group Hotels · Zante, Greece

    Applied data analytics in Excel to room occupancy, pricing trends and customer booking patterns, to optimise resource allocation.

  6. Sep 2021 – Oct 2022

    Civil Engineering Teaching Fellow

    School of Civil and Environmental Engineering, The University of Edinburgh

    Lectured, designed course material and was responsible for the academic examination of two courses: Environmental Engineering and Water Engineering. Delivered classes in Engineering Mathematics (vector calculus and statistics) and Engineering Design, and did scientific programming for wave simulations with Python and finite-element methods.

    Panos lecturing on plug-flow stream modelling, with equations on the slide
    One of my most enjoyable times: teaching engineering at the University of Edinburgh.
    A lecture slide designed by Panos
    From back when I felt like the god of editing presentations.

Education

  1. Jul 2017 – Jul 2021

    PhD, Materials Science

    The University of Edinburgh · School of Chemistry (Prof. Neil McKeown)

    Thesis: Design and synthesis of porous adsorbents for water purification.

    • Applied regression methods and data modelling using MATLAB® and Origin®.
    • Used computational techniques to fit physisorption and chemisorption models, and investigated how linearisation affects the statistical error distributions in regression.
    • Two nominations for the "Tutor of the Year" award.
    Panos at work in the lab during his PhD
    Captured in action, performing "cutting-edge research".
  2. Sep 2013 – Jul 2017

    BSc, Chemistry First Class

    The University of Edinburgh · School of Chemistry (Prof. Michael Shaver)

    First-class honours. Achieved the highest average (GPA) in my year and received the award for Outstanding Academic Performance.

What I learnt during my STEM education

  • Statistics & modellingStatistical error analysis and error functions to quantify and select the best models, with linear and non-linear regression.
  • Materials synthesisSynthesis of soft-matter materials, colloids and various chemical entities for diverse applications.
  • NMR spectroscopyStructure elucidation and compound verification, including solid-state NMR.
  • Chromatography & mass specGC-MS and LC-MS for the analysis of complex mixtures.
  • IR spectroscopyFunctional-group identification and compound analysis.
  • UV-vis & fluorescenceQuantitative analysis of compounds and tracing of free radicals in various systems.
  • ICP-OESPrecise measurement of trace elements in diverse matrices.
  • Air-sensitive techniquesSchlenk lines and glove boxes for controlled-environment experiments.
  • Materials characterisationDetailed characterisation to confirm compound integrity and function in specific applications.