AI In Digital Worlds - Computer Use
Aug 15, 2025I think language models, as simulators of human behavior, should be given the same tools as humans to interact with the digital world we share.
While AI is often associated with digital applications, its role in hardware and physical environments is just as exciting! To explore AI past software, I built and trained a robotic arm capable of learning and replicating tasks using a machine learning model called an Action Chunking Transformer (ACT). To do this, I 3D printed and wired both a “leader” and a “follower” arm, allowing me to teleoperate the follower and record demonstrations such as placing blocks into different buckets. Using this dataset, I trained the ACT model to learn from my examples, enabling the robotic arm to autonomously perform the task and even generalize to fixing its own mistakes.
Cisco
Part of the MarTech Portfolio & Innovation Team managing and integrating innovative marketing technology into our tech stack. Technical SME for GenAI solutions, hands-on development and internal consulting for platform-driven and internally-built AI software.
The DTH Media Corp.
Led and trained an advertising sales team of 15 reps. Designed and implemented the commission model, training program, and various new advertising products. Worked with local and national clients as the end-to-end sales and fulfillment rep. Top performing rep for a year and a half straight. Reported to a board of directors.
Degree in Economics, Statistics & Information Systems
Are you prompting the model or is the model prompting you? rewrAIt offers a unique 'text-editor' style interface for turn-based conversations with large language models that lets you revise any part of an LLM conversation- system, user, or AI messages, at any point in time. Add, modify, and remove context or change providers to your liking. Control the narrative with AI before it controls you.
seb-ocr is a custom built OCR and entity extraction pipeline for processing unstructured historical documents- part of a larger academic research project in the field of political science. Relying on LLMs to transcribe large volumes of scanned documents from official archives, then performing named entity recognition and extraction for downstream analysis.
QuicKB is an end-to-end machine learning pipeline that turns unstructured text into optimized, semantic knowledge bases with complimentary finetuned embedding models ready for RAG/AI retrieval. It combines the latest chunking approaches, synthetic training data generation, and dimension-reduced embedding model finetuning to create personalized domain-specific retrieval systems that are both more accurate and more efficient than generic methods.
NeedleInAVidStack is a lightweight streamlit app that rapidly identifies, timestamps, and extracts specific content across large video and audio libraries. Rather than the tedious process of manually scrubbing through video and audio files yourself, NeedleInAVidStack uses Google's Gemini AI models to automatically and efficiently parse out exactly what you're looking for in a fraction of the time.
ppt2desc converts PowerPoint presentations into comprehensive machine-readable text formats using vision language models. It captures the full semantic meaning of slides by interpreting how text, graphics, and charts relate to each other- a crucial part of presentations that traditional scraping tools miss. Compatible with major AI platforms including OpenAI, Gemini, Anthropic, GCP Vertex AI, AWS Bedrock, and Azure AI Foundry.
I think language models, as simulators of human behavior, should be given the same tools as humans to interact with the digital world we share.
Need to get in touch? Contact me here