HR Policy Clustering and Similarity Search Tool: Designed for processing and analyzing HR policy documents in Greek. Capabilities: text chunking, embedding generation, clustering, and semantic similarity detection, with the Qdrant vector database for data management. For identifying overlaps in policy content and generating recommendations.

Executed Fine-tuning & Benchmarking, optimizing 12+ LLMs (Gemma-family, Mistral, LLaMA, etc) across 6+ datasets (GSM8K, BoolQ, IMDB, Alpaca-GPT4 & more). Delivered a research-level contribution—model training, evaluation, insights, benchmark comparisons & documentation. Also crafting a custom dataset from open-sourced V0 system prompts.🛰

A neural network-based text generation model using a vanilla MLP architecture to predict the next character in a sequence. The project includes a Streamlit app for customizing parameters like embedding size and block size, enabling real-time experimentation and optimization.

A comprehensive end-to-end project demonstrating t-Distributed Stochastic Neighbor Embedding (t-SNE) for visualizing high-dimensional data using the MNIST handwritten digits dataset. Includes data preprocessing, parameter exploration, and insightful visualizations.

🛍️ An interactive Streamlit app to explore, cluster, and generate smart recommendations on Amazon products. Includes preprocessing, KMeans clustering, visualizations (boxplots, t-SNE), and a data-driven suggestion system for sellers.

VibeMap is an open-sourced sentiment-aware tweet visualizer that performs a scalable sentiment analysis tool that visualizes tweet embeddings in 3D using t-SNE and emotional clustering.

Scan your data and get useful insights