umap-hdbscan

Behavioral segmentation of open field in DeepLabCut, or B-SOID ("B-side"), is a pipeline that pairs unsupervised pattern recognition with supervised classification to achieve fast predictions of behaviors that are not predefined by users.

Browser-first text exploration, clustering, and semantic search. Use AI to analyze, search, and chat with your private documents without ever uploading them to the internet - all running on your own device.

Filtering the stockmarket into clusters with financial ratios built on quarterly snapshots over 5 years of data. The goal was to see the evolution paths of companies over time and find clusters that develop future winners. Methods used, PCA, UMAP, HDBSCAN, KMeans.

Developed an unsupervised pipeline that clusters financial headlines into 100+ meaningful topics using BERTopic and HDBSCAN. Integrated sentiment analysis to track market sentiment volatility and visualize topic trends across stock tickers and time. Combined multiple datasets to create a scalable, finance-ready insight system

Implemented machine learning across HR, Sales, Marketing, and PR to improve decision-making. Used models like XGBoost, Prophet, LSTM, clustering, and NLP to enhance retention, forecasting, segmentation, and sentiment analysis for business growth.

Repository for fine-tuning and clustering sentence embeddings for Food Items

User Clustering Pipelines with BERT Models on Long and Heterogeneous Tweets - BSc Thesis

DeepSeek & Public Opinion

Celem projektu było wykonanie analizy budżetów partycypacyjnych polskich miast pod kątem istniejących w nich wspólnych tematów, trendów, zależności.

A Python CLI for document clustering using SentenceTransformers embeddings, UMAP dimensionality reduction, and KMeans or HDBSCAN.

Document Clustering, Summarisation and Visualisation on 20NewsGroup

AI-powered customer reviews intelligence platform that combines RoBERTa, UMAP, clustering, and LLMs to turn customer feedback into product insights and actionable business recommendations.

An app that displays a daily-updated map on crime & crime hotspots/clusters in Philly.

Group project for the NUS module "IT1244 Artificial Intelligence: Technology and Impact"

CUBE v5 (Comprehensive Unsupervised Behavioural Explorer) is a point-and-click pipeline for unsupervised animal behaviour discovery from DeepLabCut pose, with UMAP, HDBSCAN, 2D and multi-camera 3D support

User Clustering Pipelines with BERT Models on Long and Heterogeneous Tweets - BSc Thesis

End-to-end NLP pipeline for news classification, semantic clustering, and automated newsletter generation using UMAP, HDBSCAN, and LLM

Adapted BERTopic pipeline for Topic Modeling the arXiv dataset

By utilizing various features, the Project assists the organization in categorizing countries and determining which country the NGO should allocate funds to.

AI-powered research gap discovery engine that mines scientific literature, builds citation graphs, clusters research domains, and identifies under-explored connections using SPECTER2, Graph Neural Networks, and semantic analysis.