Andreas Hiropedi

This web-app was developed as part of my internship at nihito, and this tool is currently available on the nihito platform (https://platform.nihito.io). The app allows the user to create a to-do list by simply typing in their tasks in natural language, and then uses a large language model to extract the key information from the user's input, such as the task description, the deadline for each task, and any other relevant details. The app then organizes the tasks into a structured format and displays them in a user-friendly interface, allowing the user to easily manage their to-do list and keep track of their tasks and deadlines.

This web-app was developed as part of my internship at nihito, and this tool is currently available on the nihito platform (https://platform.nihito.io). The app allows the user to upload a dataset in csv format, and then uses a large language model to extract useful insights from the data, generate visualization plots, and also create a detailed report explaining the key features of the dataset, such as the distribution of values for each feature, the correlation between different features, etc.

This simple web-app was developed as part of my internship at nihito, and this tool is currently available on the nihito platform (https://platform.nihito.io). The app uses the OpenWeatherMap API and allows the user to first select any location around the world by typing it into a search bar and looking and the matches found (if any), and then look at a detailed breakdown as well as visualisations of the weather conditions at the selected location.

As part of the Data Science Lab course at ETH, together with my groupmate, we performed an in-depth analysis of N2O emissions at different agriculture sites throughout Switzerland, which included cropland, grassland or a mix of the two (cropland with a temporary period of grassland).

Once we finished our analysis, we augmented our datasets by computing additional predictors that are commonly used in literature, and we then conducted a series of experiments using several different machine learning algorithms (some of which we also applied hyperparameter optimisation to) and explored their ability to predict N2O emissions at each of these individual sites.

I developed this GenAI summarizer as part of my internship with nihito, and this tool is currently available on the nihito platform (https://platform.nihito.io). The project involved working on a tool that scans your mailbox for newsletter e mails (read or unread) from the past week, extracts all this information, feeds it into a large language model (LLM) using the OpenAI API in order to generate summaries for each of the found newsletters, and then sends an email response back to the user with all the generated summaries in a nice format.

This tool was designed as a web app that is linked to a SQLite database to allow for multi-user access and also for easy re-access and setting up automated pipelines to execute the fetch-summarize-return process at set intervals of time (weekly).

The idea came to me once I realised that GitHub may not be the most creative way of showing off my skills and past projects, and therefore a more interactive option would be better. I developed this website using SvelteKit, and leveraged HTML and CSS for additional styling of components.

As part of the CS50 SQL course final project, I developed a mock database that aims to mimic the kind of data that a service like Flight Radar would store in order to display information about flights, albeit on a much smaller scale. The schema for the database, as well as the set of queries, were developed and tested using SQLite.

As part of a group project for the Machine Learning Practical course at Edinburgh University, I worked with two other students on experimenting with several deep learning approaches for detecting implicit hate speech in tweets. Our approach used a three stage classification, first differentiating between hate/ no hate comments, then for hate comments differentiating explicit from implicit hate, and for implicit hate comments classifying them into several categories. Our hybrid approach, combining a transformer, namely BERT, with a convolutional neural network (CNN) showed very promising results on our chosen dataset (which we also augmented using back-translation).

An in-depth analysis of the performance of the NHS A&E services before, during, and after the Covid-19 pandemic. This was a group project on which I worked with two other students, and together we performed several statistical analysis such as K-means clustering and generated multiple visualisations to answer a series of questions.

An altered version of the game of Connect-Four, developed using Java. The users can select the dimensions of the board as well as the number of dots connected that is needed to win, and they can either choose the multi-player mode (playing against another human) or to the single player mode, where they can choose the difficulty of the AI they are playing against.

As a continuation of my IB project which I started developing in high school, I created a web dashboard (using Python and Streamlit) to display graphs and values of cryptocurrency predictions for Bitcoin, Ethereal and Dogecoin. The user gets to select the date in the future up until which the predictions should be generated, and the dashboard then uses all past data (from Yahoo finance) to generate predictions up until the inputted date.