Hello, I'm a research assistant at National University of Singapore doing research on large-scale fact-checking, information retrieval and natural language inference. Previously, I was at Plural AI, where I worked on entity extraction from webpages and semantic search, and at Donaco, where I worked on contextual recommendation of charity campaigns. I'm passionate about AI research, improving the way we discover and organize information.
Working on new approaches for large-scale automated fact-checking, using weakly-supervised latent retrieval and natural language inference models.
Plural AI is a fintech startup backed by EF, Speedinvest, AI Seed (and more), that builds a knowledge engine for the finance industry. The company focuses on extracting content from corporate websites and financial statements and on question answering to complex financial queries (ex. construction companies in London with turnover growth > 50%).
At Plural AI, I developed a novel state-of-the-art concepts extraction model to extract products, industries, brands and business activities from corporate websites. Experimented with all the modern approaches, pre-trained language models (BERT, ELMo, USE), word2vec, Transformers for sequence labeling, independently recurrent neural networks, self-attention.
I owned the whole concepts extraction pipeline, from the annotation tool development and website crawling, to the productionization and deployment. Extracted concepts from more than 800K websites, and more than 80M text blocks. Improved the span based website level F1 score of the initial unsupervised heuristic approach by more than 25%.
Also, I developed a novel scalable semantic search service of companies using an autoencoder for dimensionality reduction and a fast approximate nearest neighbors index.
Worked on the advertising platform and modeled real-time user behavior using machine learning for better ad serving.
Designed a content-based recommendation system for recommending relevant charity campaigns to the content of news articles.
Responsible for the redevelopment of HotBot metasearch engine.
Worked as a freelance software engineer on metasearch engines for various companies and on many other small personal projects like social networks, news aggregators and computer vision software.
Fun fact: A small piece of custom search engine software that I developed and sold for $15 online, was also used by a company in Turkey (Geliyoo) that got €2.500.000 funding from the government. It became a quite popular story in Turkey for a while!
Thesis: WAMBy: Web-based open domain QA system using information retrieval techniques and machine learning (pdf in Greek)
C, C++, LaTeX, Matlab, Game Maker Studio, Video Editing
Skiing, Table Tennis, FPV Drone Racing, Piano
As a computer science researcher I am interested in machine learning and mainly in applications in natural language processing and understanding. I am also interested in information retrieval and recommender systems.
As a software engineer I hate working on already existing projects without making an impact. I like working on novel ideas from scratch that can change the world and improve our daily life.
As a human I like travelling, tasting food and meeting smart people with great ideas.
Large-scale semantic search using neural ranking models.
Personalized news recommendation and summarization.
Web-based open domain QA system using information retrieval techniques and machine learning.
A platform that connects elderly with their relatives and paid carers when they fall down, get lost or need help using Wear OS smartwatches. Won the first place in the 1st Bioengineering Hackathon in Greece.
The world's first open source torrent search engine.
Maze generator, editor & solver in Java.
Developed in Hackathon Thessaloniki, an online responsive web application for instant issues reporting in the city.
Developed a metasearch engine that combines results from Google and Bing (personal contract).
A chatbot I developed for the The Chatterbox Challenge 2014.