Thanos Pantos

Building agentic AI tools for scientific research spaning from RAG and information extraction to multi-agent research workflows.

Tech: Python, LangGraph, LangChain, Databricks, AWS

Taught in 3 training programs (approx. 60 participants) in collaboration with ReGeneration, Project Future, Papastratos, Pfizer and NBG. Target groups include young graduates and employees. Main areas of focus: python, data science, machine learning

- Developed an end-to-end machine learning solution for the virtual assistant chatbot in the Piraeus Bank mobile app. Utilized ADA embeddings and BERTopic for cluster creation to identify frequently asked user questions. Leveraged prompts for theme creation and summarization from representative topics and keywords within the formed clusters, ensuring effective topic modeling and classification.
-

- Developed and implemented a data analysis system for identifying and eliminating
outliers, resulting in a 95% reduction in the number of samples required for
debugging. Additionally, successfully deployed a web-based application utilizing the
FastAPI framework and Docker containerization technology.
- Utilized the Apriori algorithm for mining association rules and detecting changes in
customer b

- Implemented an automated pipeline for training and evaluating a wide array of
regression models on large datasets using PyCaret. Adroitly managed
hyperparameters utilizing Hydra and rendered the pipeline adaptable through the
utilization of YAML in Python.

Tech stack:
Python, R, Shiny, Julia, Hive, Bash, Git, PySpark

- Developed a sophisticated heuristic model to classify and annotate cargo types for
various vessel classes by effectively synthesizing multiple sources of data and
incorporating business-specific heuristics. Achieved a commendable f1 score of 0.9.
- Conceptualized and implemented several python scripts to monitor and analyze the
performance of the platform. Regularly updated an Influx database vi

- Developed and implemented Python-based web crawlers to collect data from various
online platforms such as blogs and forums. The collected comments were
subsequently processed through advanced techniques of data cleaning and were fed
to an Artificial Intelligence algorithm for the purpose of analysing user behaviour
characteristics.
- Employed machine learning methodologies to construct regressio

Tech stack:
R, Python, Tableau, MySQL, Git

Apr 2024 · Expires May 2027

Apr 2024

Aug 2023 · Expired Aug 2024

Jan 2023

https://dsml.ece.ntua.gr/

Thesis (https://dspace.lib.ntua.gr/xmlui/handle/123456789/56490) : Reserve volume estimation of oil storage tanks based on remote sensing images.
The objective of this thesis is to develop a deep learning-based method utilizing YOLOv3 and shadow extraction algorithms for estimating oil tank volumes from high-resolution satellite imagery, with the goal of fulfilling the r

Sep 2021

Sep 2021

Jan 2020

Relevant Modules:
• Econometrics
• Mathematics
• Linear Algebra
• Statistics
• Probability Theory
• Time-series Analysis
• Applied Mathematics in Economics
• Financial Mathematics