STA Presentation Website

Started: 04/04/2024

Description: This website is the website showcasing the livestream for people online to watch!

Thank you to Nicolas for amazing banner.

Web Editor Help

Started: 04/04/2024

Description: This project helps staff at COLA to search easily the right documentation with the use of Retrieval Augmented Generation. Currently, they use Google Dialogue when it only returned one matching result and when adding a new documentation, people need to add a lot of phrases to the document in hope that it can match these phrases with user’s search. Moreover, the model or the techniques they use with Google Dialogue is a black box so it is hard to improve the system. Therefore, I come in to help with upgrading the retrieval part of the system with using Cosine Similarity Search when I converted the documents into vector embedding using Sentence Embedding Model. Then, all of the retrieved relevant documentaries are fed into LLM so that it can generate helpful responses.

All of the functions that wrote:

__construct()

calling_hugging_face($text)

convertHelpCorpusAndStoreToDatabase()

storeEmbeddingIntoDatabase(array $embeddingVector, $path, $title)

processEmbedding(Request $request)

return_best_match_pgvector($embeddingData, $k = 5)

show_validation()

showForm()

The results:

Without intent:
[2025-03-25 21:13:03] Validation Top K: 1, Validation Score: 61, Total: 134 -> 45.5%

[2025-03-25 21:11:49] Validation Top K: 2, Validation Score: 82, Total: 134 -> 61.2%

With intent:
[2025-04-04 19:25:30] Validation Top K: 1, Validation Score: 73, Total: 134 -> 54.5%

[2025-04-04 19:26:44] Validation Top K: 2, Validation Score: 94, Total: 134 -> 70.1%

After eliminating not useful validation entry:

[2025-04-04 19:41:12] Validation Top K: 2, Validation Score: 92, Total: 126 -> 73%

[2025-04-04 19:42:55] Validation Top K: 1, Validation Score: 72, Total: 126 -> 57%

[2025-04-04 19:51:38] Validation Top K: 2, Validation Score: 92, Total: 126
[2025-04-04 19:53:53] Validation Top K: 2, Validation Score: 92, Total: 126

Feb – March Update

Project: UT Instapoll

Started on: Feb 20, 2024

Staff Guidance: Chris Pittman

Description: In this project, I redesigned and implemented the look of the application. It looks much better now on the website, but I still need to work on getting this look onto the pdf. Currently, the pdf does not have the logo and every question appears on different page.

Before

After

PDF:

Project: Web Editor Help

Started on: Feb 24, 2024

Staff Guidance: Geoff Boyd

Description: This project involves designing a better retrieval system for the web editor of COLA website. The website serves staff member to search for information easily with the documentation in the database. There are a lot of steps involved with redesigning the retrieval step, starting with implementing the pgvector database, implementing sentence-transformers for converting text to embedding, storing these embeddings into the database and then using cosine similarity to match.

Currently, I have done the first step of establishing the pgvector database and input some fake data into it. The first stage takes me a lot of time to set up and transition into pgvector from sql. But it has been good for now. The next stage I am doing is how to make sentence embedding to convert text to vectors for me.

Project: UT Instapoll

Started on: Jan 27th, 2024

Staff Guidance: Chris Pittman

Description: The aim of this project is to incorporate the PDF integration into UT Instapoll. This features help student to download all of the poll responses as a study guide for a test. The PDF study guide comprises all of the questions, answers and the correct answer choices.

This is the continuation of part 1. In part 2, I implemented the library to generate PDF, customize the PDF, let user download the PDF to their machines, design the PDF and implement it.

• Implemented the library to generate PDF

• Design Figma

• Implement the design

Project 1: Polls

Started on: Jan 27th, 2024

Staff Guidance: Chris Pittman

Description: The aim of this project is to incorporate the PDF integration into UT Instapoll. This features help student to download all of the poll responses as a study guide for a test. The PDF study guide comprises all of the questions, answers and the correct answer choices.

Documenting my thought process for this project. There will be two parts:

1. Creating the route, the controller, the blade template.
2. Transform the Blade Template into PDF

1. Create the route, the controller and the blade template:

Route: This step establishes the URL end point to let the user access to get the PDF. For this, I chose teacher/course/1/poll_pdf.

Controller: The controller part controls the logic of back-end behind this operation. It will get the polls from the database so that I can access the question, response and the correct answer choices.

Blade Template: The blade template displays the HTML with all of the poll responses and correct responses. The template for the PDF is still in consideration.

HTML Result: The result looks something like this.

2. Part 2 is still in progress

This part, I still need to integrate the library to get PDF integration working.

Project 2: Qualtrics-LTI

Started on: Jan 20th, 2024

Finished on: Jan 27th, 2024

Staff Guidance: Chris Pittman

Description: Adding key and value to facilitate the process of adding these values onto the page.

November Updates

November went by so quickly. It has been half of the month. For this past month, I have done more work with Shiny App and Texas Politics project.

Shiny App Project

Advisor: Stacy Vlasits

Description: For this project, I continued to work on implementing features for the application built in Shiny App. The client requested two more features to be added: working with URL input and working with CSV files.

In tackling these two challenges, I isolate the function that handle the data uploads. The answer is that it is in page 2 and then, I looked at how to handle URL input and turn that into a table format. When I looked at the function that handles uploads of the data, I actually find a solution that works for both URL input and CSV input. I just need to dynamically change the extension accordingly to whatever the format of the file uploads. For example, the CSV would have csv extension and TSV would have tsv extension. Moreover, for handling URL string, I just need to extract the base of the string to get the extension.

A function like this solves both problem at the same time:

# Get file extension from either uploaded file or URL
ext <- if(!is.null(input$file1)){
tools::file_ext(input$file1$name)
}else if (!is.null(input$urlInput) && input$urlInput != “”){
# extract extension from URL
tools::file_ext(basename(input$urlInput))
}else{
“csv”
}

Texas Politics

Advisors: Chris Pittman

Description: For this project, I only get the overview of what the project is about. I did not start to look closely into the solutions. From the outset, it seems like the project involves a place for storing all of the updated elements including charts and blogs. Also, the project incorporates a change in the database with a boolean value “updated me” so that the blog, chart will be included into that place. The idea behind is that every year, there would be blogs, charts and other elements that get to be updated. Currently, there is one guy that have all these information but to streamline the process, we can leverage technology to keep track of all these information. It helps to onboard new people quickly to the updated part of the project.