Articles/Projects:
genui.com
GraphRAG vs. Baseline RAG: Solving Multi-Hop Reasoning in LLMs
In this article, I compare GraphRAG to baseline RAG, showing how GraphRAG’s use of knowledge graphs enables better multi-hop reasoning across documents for more accurate results.
genui.com
Building a custom RAG pipeline with LlamaIndex and Llama3.2
I walk through building a custom Retrieval-Augmented Generation (RAG) pipeline with LlamaIndex, Llama 3.2, and LlamaParse to improve LLM accuracy with external data, covering ingestion, embedding, querying, and future optimizations.
genui.com
Building your own software development team with chatGPT and AutoGen
I explore how AI’s untapped potential can be unlocked through multi-agent collaboration, where multiple LLMs work together to enhance efficiency and innovation in software development and beyond.
genui.com
Do Larger Context Windows Remove the Need for RAG?
I examine whether the growing context windows in AI models could eliminate the need for Retrieval-Augmented Generation (RAG) by including all necessary context directly in prompts, and what this means for performance and cost.
github.com
Gunshot sound detection in urban environments
I implemented a double-headed EffNet with PyTorch to detect gunshot sounds using spectrograms and chromagrams.
stanleyjovel.com
Comparison of AI Models for Celestial Body Classification
I presented this study at the 2nd Central American Space Congress, comparing two AI approaches for classifying stars, galaxies, and quasars. Content is in Spanish.
github.com
Mapping Urban Dynamics: Crime, Property, Subway Ridership, and Business in NYC
I investigated relationships between crime rates, property values, subway ridership, and business perceptions using big data technologies like Zeppelin notebooks, Spark SQL, Presto, and Tableau.
stanley-jovel.github.io
Self-Driving Car Learning with Behavior Cloning and PPO in Godot
I explored behavior cloning to reduce training time for a PPO agent mastering self-driving car maneuvers in Godot.
github.com
Machine Learning, Deep Learning Training and Inference
I trained a CNN on the GTSRB dataset with PyTorch (98% accuracy) and worked on ML projects like stock market and Olympics predictions.
dev.to
React Hooks Simplify Controlled Components
I explain how React hooks simplify building controlled input components.