createRAGstore
createRAGstore is a set of custom python scripts that can scrape a public website for information and create a vectore store to utilize Retrieval Augmented Generation with a pretrained Large Language Model, enabling users to create a chatbot that is a subject matter expert on the content of said website.
I developed this set of scripts to gather sets of rules for ScryNet, which is a chatbot built with llama3.1 that is a tabletop role-playing game rules expert, game master assistant, non-player character generator, dialog generator, etc.
createRAGstore has multiple steps and checkpointing built in, as the website I used to test it has over 65000 unique URLs. This will allow users to pause their information gathering process, and is a failsafe protecting against loss of power, internet connectivity, or site connectivity.
The most difficult part of development for this project was generalizing the process so that it can be used to create a subject mattter expert from any website, instead of just the one I needed it for. It is available on my Github page, which can be reached on this site under the "Contact" tab.
SayWhat
SayWhat is a multi-platform, multi-media translation application developed as a capstone project for my Bachelor's of Science in Computer Science. Created collaboratively with two other developers over a three-month period, the application successfully delivers comprehensive translation capabilities across multiple formats:
• Live recorded audio translation with speech output
• Text extraction and translation from both live-captured and stored photos
• Live typed text translation
• Desktop Mode enabling real-time audio-to-text translation between two users in their
preferred languages
The project demonstrates strong technical implementation skills through various API integrations and cross-platform development, resulting in a fully functional mobile application compatible with both Android and iOS. The development process required significant time management, research, and collaborative teamwork to successfully deliver this versatile translation solution within the three-month timeframe.
Here's a walkthrough video of the application.
Mental Health Chatbot
I developed a Mental Health Chatbot LLM as an independent one-month project using the chatterbot Python library. I trained the bot on a comprehensive dataset obtained from Huggingface with the purpose of creating a chatbot that could provide encouraging advice to users struggling with their mental health, while recommending professional help when deemed necessary.
During development, I faced several challenges including:
•Finding a large enough dataset for training the model
•Cleaning the data multiple times to reduce hallucinations in responses
•Building a web-based user interface for the first time
•Successfully deploying the web application on Heroku
•Completing all aspects of the project within a limited one-month timeframe
Overall, this project was a significant challenge that I look back on fondly. It allowed me to combine my technical skills in machine learning, data processing, and web development while creating a solution with potential positive impact on users' wellbeing.
Here is a video of some Think Aloud testing that was performed during the development process.
Whisay
I independently developed Whisay, focusing on building my own Optical Character Recognition (OCR) system to extract and translate text from images.
I completed this project within the challenging timeframe of just one week.
The application performed well when processing stored images but faced challenges with live-captured photos. To test the mobile functionality, I deployed the web application on Heroku to enable testing from my mobile device. However, I discovered that images were being compressed during transmission, which reduced the OCR system's ability to accurately recognize characters.
The primary challenge for this project was the tight one-week time constraint, which prevented me from resolving the live-capture functionality issues. Despite this limitation, I was impressed with how much functionality I was able to implement in such a short period, successfully creating a working OCR system that could effectively process and translate text from stored images.
Here is an example of how the application works. It has been simplified to run as a client-side application on this website. Upload an image containing text in a language other than English, and the application will translate it into English for you and read it aloud.
Extracted Text:
Translated Text:
KrashBox Version 1
I built a custom arcade gaming cabinet for my friend's at-home bar.
The cabinet uses two USB converters for the control arrays that are connected to a Raspberry PI 4b.
The operating system is called RecalBox, which runs a set of emulators for classic systems including NES, Sega Genesis, Atari, and many classic arcade games.
The most difficult part of this project was deciding on the design. Once I started cutting the wood everything else fell into place though. I decided to go with a punk/DIY look for the paint, and left plenty of room for the owner to add stickers of their choosing.
