• Generative Matchmaking

    This post details an AI-based matchmaking simulation for dating, leveraging vector-based semantic search for user compatibility recommendations and retrieval-augmented generation (RAG) for co-personalized first-date ideation using OpenAI’s GPT-4, OpenAI’s DALL·E-2, and Chroma, an AI-native open-source vector database. My brother, Dimitri Deychakiwsky, and I put this together over several weekends.

  • Open Foundation AI Models - Meta's Llama2-7B-Chat

    This post shares practical learnings from experimenting with Meta’s Llama-2-7B-Chat LLM via HuggingFace APIs quantized to FP16 on a 16 CPU CORE, 60GB CPU MEM, and 16GB GPU MEM hardware instance.

  • Matrix Factorization Sparsity

    This post explores how implicit user-item interaction sparsity poses a challenge for matrix factorization (model-based) collaborative filtering recommender systems.

  • Low Energy Audio Segments

    This post outlines a simple algorithm for identifying low energy audio segments. This post assumes basic knowledge of digital signal processing (DSP).

  • Ordinary Least Squares is Orthogonal Projection

    This post visualizes the equivalence of two perspectives on estimating the unknown parameters in a simple linear regression model, ordinary least squares (OLS) and orthogonal projection (OP).

  • Normalized Cross-Entropy

    This post explores a normalized version of binary cross-entropy loss in attempt to remove the effect of the prior (class imbalance within the dataset) on the resulting value.

  • About Me

    This is my first post. Here’s a bit about who I am and why I started this blog :).