search

vinnieOrdobas / ci_national_ai

0 stars 1 forks source link

readme

HomeIE - the AI Assistant that helps you to find a secure home - AI here for good!

Responsive Image Next steps - insert hyperlink to bottom of README

Introduction

-What the project hopes to accomplish.

The project aims to identify a problem affecting modern society and to provide an AI-driven solution through a website designed to address that problem. Evidencing that AI is here for good.

Problem Identification
The pain point we have identified is the lack of easily accessible, comprehensive information and assistance available online for people looking to securely purchase a home. The solutions are often buried deep within dense documents, hidden in hyperlinks in small print on webpage footers. Moreover, the information is scattered across unconnected websites, creating a tedious obstacle course for users.

Proposed Solution
Our solution, ‘Homie,’ consolidates all relevant information in a fun and engaging way through an accessible chatbot interface. We address this problem by developing a web application featuring an interactive form and chatbot. Using Chat GPT-4.0, integrated with Gemini, we responsibly sourced extensive datasets to empower the AI with the ability to offer informed and personalized advice. This service draws on a variety of data sources to consider the user’s financial situation and suggests multiple paths to homeownership or rental, including the identification of potential grants and financial assistance programs.

Implementation

Given the time constraints and the varying experience levels within our team, we assigned a Scrum Master and adhered to the Agile methodology. We utilized a Kanban board for task management, shared Excel files and Google Drive for collaboration, and Slack for communication. Daily stand-ups were held to ensure alignment, and tasks were carefully divided and assigned based on skillsets.

From the outset, we focused on ensuring the effectiveness, innovation, and practical utility of our solution. We incorporated diverse datasets to enhance accuracy and reduce the risk of AI hallucinations, ensuring that our chatbot delivers clear and reliable advice. Functionality was prioritized, with the application designed to be fully operational for a live demo.

Creativity and originality were central to our approach, and we implemented a rigorous technical process. By setting up a host coding page, establishing a review mechanism, and using wireframes and system architecture diagrams, we ensured that our project was technically sound and well-coordinated.

Process for Setting Up a Collaborative Environment on GitHub (Step-by-Step Process)

1. Setting Up the Repository

2. Main Branch Management

3. Branching Strategy

4. Push and Pull Requests

Description of our team, skillset and main contribution to the project:

Technical Details

Technology Stack

Significant Libraries Used

Additional Notes

Environment Setup and Dependencies

To set up the project locally, follow these steps:

  1. Clone the Repository: 
    
git clone git@github.com:vinnieOrdobas/ci_national_ai.git
cd ci_national_ai

User Instructions

Installation and Setup

Follow these steps to set up the project locally on your machine:

  1. Clone the Repository:
    • Use the following command to clone the project repository from GitHub: 
      
git clone git@github.com:vinnieOrdobas/ci_national_ai.git
cd ci_national_ai

Brainstorm & Research

Our team conducted a thorough brainstorming session to determine the most effective type of web application to address the identified problem. We considered various ideas and ultimately decided on a solution that leverages AI to provide accessible housing advice.

Wireframes

To visualize our ideas, we created detailed wireframes that outline the structure and design of the web application. These wireframes guided our development process and ensured a user-friendly interface.

Project Architecture

We engaged in discussions to finalize the architecture of the project. This included decisions on how to compile and integrate all components, with a particular focus on the integration of AI. Below is the architectural design we developed to serve as the foundation for the project:

Planning & Organization

Following extensive planning meetings, Deirdre was elected as the Scrum Master. She was instrumental in organizing our efforts, including setting up a detailed Kanban board that allowed us to manage tasks effectively and stay on schedule.

Data Considerations for Model Fine-Tuning with Vertex AI

When determining the data to gather for fine-tuning our model using Vertex AI, we evaluated several fine-tuning methods:

Supervised Fine-Tuning:
This approach involves using a dataset where each example includes both an input (such as text or an image) and a corresponding desired output (such as a classification label or a generated text response). Supervised fine-tuning is particularly suited for tasks such as classification, summarization, reasoning, and question-answering.

Reinforcement Learning from Human Feedback (RLHF):
This method involves fine-tuning the model based on human feedback to adjust the tone and style of the model's responses. An example of this process is when models like ChatGPT present multiple responses and ask users to rate them. While this technique can be valuable for refining the model's output style, we have determined that it is not necessary for our current objectives.

Model Distillation:
Model distillation is typically employed to transfer the knowledge from a larger model to a smaller one, creating a more lightweight version suitable for deployment on devices with limited resources, such as Android. However, this method is not relevant to our current needs and will not be utilized.

Data Considerations for Supervised Learning

Given that our training will focus on supervised learning tasks, it was crucial to carefully curate the dataset. The data had to be structured effectively, particularly in a question-and-answer format, to ensure that the model performed optimally in the targeted applications.

We recognized the need for data that could be formatted in a conversational question/answer format to facilitate supervised fine-tuning.

JSONL on Chat GPT Fine-tuning on Vertex AI

Technical Implementation and Pipeline Development

Pipeline Development: Vinnie outlined the process for creating a test pipeline, which involves containerizing the application and deploying it to Google Cloud Platform (GCP). The goal is to validate the end-to-end process before scaling up to more complex tasks. This involves gathering and formatting datasets, uploading them to the GCP Console UI, and creating a comprehensive training pipeline. Given the computational demands of training a large language model (LLM), the process will require significant GPU resources.

Pipeline Steps:

  1. Define the Pipeline:

    • Objective: Set up a simple test pipeline to validate the process.
    • Components:
      • Data preprocessing
      • A simple training loop (e.g., training a small model on a small dataset)
      • Evaluation and logging

  2. Containerize the Pipeline:

    • Step 1: Write a Dockerfile
      • Objective: Containerize the pipeline for deployment on GCP.
    • Step 2: Build and Test the Docker Image
      • Commands to run:
      • docker build -t my-pipeline .
      • docker run my-pipeline
      • gcloud auth login
      • Tag and push the Docker image:
      • docker tag my-pipeline gcr.io/<your-project-id>/my-pipeline:latest
      • docker push gcr.io/<your-project-id>/my-pipeline:latest

  3. Deploy the Container to GCP:

    • Step 1: Push the Docker Image to Google Container Registry (GCR)
      • Login to GCP and use the command line to push the image.
      • gcloud auth configure-docker
      • docker push gcr.io/<your-project-id>/my-pipeline:latest
    • Step 2: Deploy to Google Cloud Run or GKE
      • For Google Cloud Run:
      • gcloud run deploy my-service --image gcr.io/<your-project-id>/my-pipeline:latest --platform managed
      • For Google Kubernetes Engine (GKE):
      • Create a GKE cluster and deploy the container.
        • gcloud container clusters create my-cluster --num-nodes=3
        • kubectl create deployment my-deployment --image=gcr.io/<your-project-id>/my-pipeline:latest
        • kubectl expose deployment my-deployment --type=LoadBalancer --port 80

  4. Test and Monitor the Pipeline on GCP:

    • Step 1: Monitor the Deployment
      • Use the GCP Console to monitor logs, performance, and resource usage.
      • gcloud logging read "resource.type=cloud_run_revision"
      • Verify that the pipeline runs as expected and outputs the desired results.

  5. Scaling Up: Training the LLM

    • Step 1: Preparing for Full-Scale Training

      • Objective: Once the test pipeline is successful, scale up by preparing the full dataset and utilizing GPUs.
      • Tasks:
      • Data Preparation: Format and preprocess datasets as needed.
      • Upload to GCP: Upload datasets to Google Cloud Storage (GCS).
        • gsutil cp /local/path/to/data.csv gs://<your-bucket-name>/data/
      • Adjust Pipeline for LLM: Modify the pipeline to train the larger model using the full dataset.
      • GPU Resources: Ensure that the GCP environment is configured with GPUs (e.g., using GKE with GPU nodes or AI Platform).

    • Step 2: Train the LLM

      • Initiate the Training: Start the training process on the full-scale model using AI Platform’s training services or by running the job on GKE with GPU support.
      • For AI Platform: gcloud ai custom-jobs create --region=us-central1 --display-name=llm-training --config=config.yaml
      • For GKE: kubectl apply -f llm-training-job.yaml

Dataset Preparation and Optimization

In the context of time and resource constraints, we capped the dataset size at 100k rows to balance efficiency and training time. This required distilling the datasets based on user stories and required capabilities, ensuring we retained underlying patterns while reducing dataset size. This approach aims to achieve over 90% efficiency while optimizing for time. we comprehensively researched relevant data, prioritised it and converted it to JSONL conversational Q&A format with context.

Dataset prioritization

ChatGPT-Assisted Scraping and Dataset Preparation

We leveraged ChatGPT to streamline the scraping and dataset preparation process. Key advantages included:

In summary, these processes set the stage for effective supervised fine-tuning of the HomeIE model, ensuring it meets the needs of the targeted user base.

FEATURES:

EXISTING FEATURES

  1. Interactive Chatbot Assistant

    • Value Provided: The interactive chatbot serves as the primary user interface, offering personalized advice and information based on user inputs. It significantly enhances user engagement by providing immediate, tailored responses.
    • Objective: To provide users with accessible, accurate, and relevant housing advice based on their specific needs and inputs.
    • Response to Testing: Comprehensive testing was conducted, focusing on response accuracy, context understanding, and user satisfaction. Details of the testing process and results are provided in the testing section below.

  2. Form

    • Value Provided: The form allows users to input specific data anonymously, which the chatbot uses to generate personalized and relevant advice. This feature enhances the accuracy and relevance of the chatbot's responses.
    • Objective: To gather essential user data that informs the chatbot’s responses, ensuring that advice is personalized and useful.
    • Response to Testing: Testing focused on data handling, anonymity, and integration with the chatbot. Results indicate that the form successfully captures user data without compromising privacy, and this data is effectively used by the chatbot.

  3. Language Abilities of the Chatbot

    • Value Provided: The chatbot supports multiple languages, making the application accessible to a broader audience, including non-native English speakers.
    • Objective: To ensure that the chatbot can communicate effectively with users in various languages, improving accessibility and inclusivity.
    • Response to Testing: Language testing involved assessing the chatbot’s ability to understand and respond in different languages. The chatbot successfully handled multiple languages, but ongoing improvements are planned to enhance fluency and cultural relevance.

  4. API Integration with Looker Studio

    • Value Provided: This feature enables the dynamic display of real-time housing market statistics and trends through visualizations created in Looker Studio. This adds a valuable layer of data-driven insights to the user experience.
    • Objective: To provide users with up-to-date visual information about the housing market, enhancing the advice given by the chatbot.
    • Response to Testing: Testing involved ensuring that the API correctly retrieves and displays data from Looker Studio. The integration was successful, with data being accurately represented in the visualizations.

FEATURES LEFT TO IMPLEMENT (Future Features)

  1. AI-Driven Screen Narration

    • Objective: To increase accessibility by enabling the AI to audibly describe on-screen content, assisting visually impaired users.
    • Impact: This feature will make the HomeIE application more inclusive, ensuring that all users can access the information provided by the chatbot.

  2. Video Prompts

    • Objective: To enhance user engagement by providing video prompts that guide users through the process of securing a home.
    • Impact: Video prompts will offer a more interactive and educational experience, helping users understand complex information more easily.

  3. EIRCODE Integration

    • Objective: To allow users to upload their EIRCODE, enabling the chatbot to search for housing options in that specific area.
    • Impact: This feature will provide more localized and relevant housing information, making the advice even more personalized.

  4. Photo Upload Capability

    • Objective: To allow users to upload photos that the chatbot can analyze to provide advice on housing conditions or potential improvements.
    • Impact: This feature will add a visual analysis component to the chatbot’s capabilities, offering more comprehensive advice.

  5. Representation of Mapped Data

    • Objective:

Hackathon Specific Criteria:

  1. Relevance to Theme

    The theme ‘AI …here for Good’ aligns perfectly with our solution, which uses AI to provide an easily accessible service on housing and grants. This empowers the public, helping them feel more in control when seeking a home, rather than feeling powerless.

    Our AI assistant, ‘Homie,’ transforms dense literature and overwhelming statistics, tables, charts, and graphs into easily accessible information. ‘Homie’ is trained to be knowledgeable across all relevant areas, prepared for every eventuality, and designed to minimize the risk of hallucinations by bridging logical gaps.

    We have harnessed AI to make housing information more accessible and understandable to society. Our chatbot, ‘Homie,’ is equipped with an extensive and varied knowledge base, enabling it to draw connections so users don’t have to.

  2. Impact & Usefulness

    The potential impact of this web application on society is substantial. In Ireland, it is all too common for individuals to be swindled out of deposits, deceived by unscrupulous landlords, or left without a home or refuge. Our web application is designed to serve all segments of society, offering tailored advice to marginalized groups such as the previously homeless, refugees and asylum seekers fleeing war-torn countries, and the elderly seeking assistive living arrangements.

    When prompted, the AI chatbot provides extensive, tailored information that addresses the unique circumstances of each user, ensuring that they receive the most relevant and accurate advice.

  3. Ethical Use of AI

    Ethical AI principles: fairness, transparency, privacy, and accountability throughout the design and implementation of our AI solution.

    Homie adheres to these ethical AI principles by ensuring fairness in decision-making processes, allowing users to interact with the web application at their discretion, without being guided or coerced into sharing Personally Identifiable Information (PPI). The AI tains transparency through clear disclaimers outlining its limitations, which are included at the bottom of each tab. Additionally, we prioritize user privacy by implementing robust data handling practices—Homie does not retain any information, and no PPI is collected or shared by the user.

    To ensure data was handled ethically with respect to user privacy and consent, users are required to expressly state their consent where applicable. They are fully informed of any limitations through the disclaimers provided. Furthermore, if continuing with the web application, we would maintain accountability by regularly reviewing and auditing our AI processes to ensure they align with these ethical standards.

  4. Technical Implementation

    Technical Implementation

Thanks to the resources provided by TechIreland, we leveraged a combination of Google Cloud Platform (GCP), OpenAI, ChatGPT 4.0, and Gemini to build and deploy our web application.

Technology Stack and Tools

Our application was developed with a robust technology stack as described earlier in the README.

  1. Feasibility & Scalability

    • Realistic Approach: Discuss the practicality of your solution, including any potential challenges in implementation. Homie is a very practical solution to a very real, omnipresent issue in modern Ireland. Gone are the days when only wealthy people could access technology—the modern world is the Digital Age. Even someone with very limited resources could access the web application in a public library.

      It will have a simple and well-navigated layout to increase user understanding.

    • Scalability: The project has immense potential to be scaled. It is very effective with a limited amount of datasets put in...could be expanded to include a housing permission form, to create a tailored lease to offer a landlord, a rental/house buying checklist tailored to the area the person wants to buy...

  2. User Experience (UX) & Design

    We believe that HomeIE offers an intuitive, responsive, and accessible user experience. The application is designed to be interactive and user-friendly, with a clear layout and consistent appearance. Key accessibility features, such as alt text and aria-labels, ensure that the website is usable for individuals relying on screen readers.

We incorporated the five planes of user experience in the design of HomeIE:

  1. Strategy Plane: Defined the objectives of the website, focusing on providing accessible housing advice.
  2. Scope Plane: Determined the functionalities required to meet user needs, such as the chatbot and API integrations.
  3. Structure Plane: Organized the content and features in a logical and user-friendly manner.
  4. Skeleton Plane: Developed wireframes and mockups to visualize the layout and navigation.
  5. Surface Plane: Applied visual design principles to ensure the website is aesthetically pleasing and consistent.

This is further discussed later in the README.

  1. Presentation & Pitch

    • What It Means: This criterion evaluates how well your team communicates the idea and solution. It looks at the clarity, conciseness, and engagement of your pitch.

    • How to Address It:

      • Clear Communication: Ensure your pitch is clear and focused on key aspects of the project. Avoid technical jargon unless necessary and explain concepts simply.
      • Engagement: Make the presentation engaging by using visuals, real-world examples, and a compelling narrative.

    • Real world examples = Need for housing advice, Irish Times article on immigration

Ethical Use of AI:

  1. Fairness & Bias Mitigation:

    • Bias Mitigation: Our project, “Homie,” is designed to provide housing advice tailored to individual needs while actively avoiding biases in data and decision-making. We ensured that the datasets used were diverse and representative of various social groups, including marginalized communities, to prevent the perpetuation of existing biases.
    • Fairness Mechanisms: We implemented fairness checks in our AI models to detect and address any potential biases. These checks help to ensure that “HomeIE” provides equitable recommendations regardless of the user’s background or situation.

  2. Transparency & Explainability:

    • Transparency: We prioritized transparency in “Homie” by including clear, user-friendly explanations of how the AI generates its recommendations. Each interaction with the chatbot is accompanied by an explanation of the data sources and reasoning behind the advice given.
    • Documentation: Our team provided comprehensive documentation that outlines the workings of the AI model, making it accessible to both technical and non-technical users. This ensures that users understand the process behind “Homie’s” decision-making.

  3. Privacy & Data Protection:

    • Data Protection: “Homie” was built with user privacy as a top priority. The chatbot does not retain any Personally Identifiable Information (PPI), and all user interactions are anonymized to protect privacy.
    • Compliance: Our project complies with data protection regulations, ensuring that any data handled is done so securely and in accordance with relevant standards. We also implemented encryption and other security measures to prevent unauthorized access.

  4. Accountability & Responsibility:

    • Accountability Measures: We established clear accountability by implementing logging and auditing mechanisms that track “HomeIE’s” interactions and decisions. This allows us to monitor for any errors or unintended consequences and correct them as needed.
    • Ethical Considerations: Our team carefully considered the ethical implications of deploying “Homie,” particularly in terms of its impact on users’ decisions regarding housing. We are committed to ongoing monitoring and updates to ensure responsible use.

  5. Social & Environmental Responsibility:

    • Social Good: “Homie” is designed to promote social good by making critical housing information accessible to all, including marginalized and vulnerable groups. This contributes to a more equitable society by empowering users to make informed decisions about their housing options.
    • Environmental Responsibility: While the primary focus of our project is on social impact, we have also considered the environmental footprint of our solution. We optimized the AI’s operations to minimize energy consumption and maximize efficiency.

  6. Inclusivity & Accessibility:

    • Inclusive Design: “HomeIE” was developed with inclusivity in mind, offering accessible interfaces and easy-to-understand language to ensure it can be used by people from all walks of life, including those with disabilities.
    • Marginalized Groups: Our team specifically tailored “HomeIE’s” advice to address the needs of marginalized groups, such as refugees, the elderly, and low-income individuals, ensuring that they receive relevant and actionable information.

National AI Strategy

How does the project align with the National AI Strategy?

This Strategy sets out how Ireland can be an international leader in using AI to benefit our economy and society, through a people-centred, ethical approach to its development, adoption and use.

In recognition of the wide-ranging effect AI will have on our lives, this Strategy considers AI from a number of perspectives. These are:

Building public trust

Leveraging AI for economic and societal benefit

Who the project is targeting?

The project is targeting Irish residents who are seeking clear, accessible advice on securing a home. This includes a wide range of individuals, from first-time homebuyers to those facing challenges such as financial constraints or limited knowledge of the housing market.

How it will be useful to a target audience.

The project will empower users to make well-informed decisions about securing a home by offering tailored advice that takes into account their unique circumstances. This advice is based on comprehensive research and the specific information they provide. What sets our solution apart is its ability to draw from a wide range of reliable sources, ensuring that users receive the most accurate and relevant guidance available.

TESTING

We have conducted extensive testing to ensure that the web application has reached an acceptable level of functionality. The various types of testing conducted are outlined below:

Manual Testing

We performed comprehensive manual testing on all features of the application to ensure they functioned as intended. This included testing each feature to verify that it could complete its designated tasks. Additionally, we manually tested the website on different devices, including phones and laptops, to ensure full responsiveness and functionality across various viewports.

Lighthouse Testing Using Developer Tools

Lighthouse testing was conducted using developer tools to assess the website's performance, accessibility, best practices, and SEO. We viewed the website on different browsers, both manually and online, to ensure a consistent and accessible experience across all platforms. Accessibility was a priority, and Lighthouse provided crucial insights that helped us refine the user experience.

Lighthouse

Validator Testing Using Online Validation Tools

To ensure that our code adheres to industry standards, we used the following validation tools:

Overall Testing Summary

In addition to the above, we manually tested all of the project's features to ensure they operated as intended. The testing process was thorough, involving various approaches to validate the functionality, responsiveness, and accessibility of the website. By interacting with developer tools and performing manual checks across different devices and viewports, we ensured that the website is fully functional and responsive.

Additional Notes

  1. Testing Across Different Viewports

    • Once deployed, the website was opened on various viewports to test its responsiveness.

    Responsiveness Testing:

    Device Responsive (<=700px) Links/URLs Images Work
    iPhone 12 Pro Yes Yes Yes
    MacBook Pro Yes Yes Yes
    iPad Air Yes Yes Yes
    Nest Hub Yes Yes Yes

    Screengrabs of Different Tests:

    • iPhone 12 Pro: iPhone 12 Pro
    • MacBook Pro: MafBook Pro
    • iPad Air: iPad Air
    • Nest Hub: Nest Hub

  2. Testing Across Different Browsers

    • The website was opened on various browsers to ensure compatibility and intended functionality.

    Browser Compatibility:

    Browser Responsive Functionality
    Chrome Yes Yes
    Firefox Yes Yes
    Safari Yes Yes
    Internet Explorer Yes Yes

Bugs and Issues Identified During Testing

Addressed Bugs and Problems:

Unaddressed Bugs and Problems:

Additional Notes

Bugs

  1. Problem: [Describe the specific issue encountered]
    • Solution: [Explain how the issue was resolved]

Deployment Process

The site was successfully deployed to GitHub Pages. The following steps outline the deployment process:

  1. Repository Selection:

    • Navigate to the relevant GitHub repository and go to the Settings tab.

  2. Branch Selection:

    • From the source section drop-down menu, select the Master branch (or Main branch depending on the repository setup).

  3. Deployment Confirmation:

    • Once the branch is selected, the page will refresh, and a ribbon display will indicate that the deployment was successful.

The live link to the deployed website can be found here: https://marine-copilot-433717-q0.ey.r.appspot.com/clientui/#info

User Experience

We incorporated the five planes of user experience in the design of HomeIE:

  1. Strategy Plane: Defined the objectives of the website, focusing on providing accessible housing advice.
  2. Scope Plane: Determined the functionalities required to meet user needs, such as the chatbot and API integrations.
  3. Structure Plane: Organized the content and features in a logical and user-friendly manner.
  4. Skeleton Plane: Developed wireframes and mockups to visualize the layout and navigation.
  5. Surface Plane: Applied visual design principles to ensure the website is aesthetically pleasing and consistent.

Collaboration Tools

Throughout the project, we used various collaboration tools to streamline development:

1. Strategy Plane

When deciding on what features to include, we compared viability and feasibility (whether the team had the skills, resources, and time to implement the desired features) against importance (alignment with business goals and user needs). The core element of this plane is making strategic decisions about what features and elements to include or exclude, ensuring that the project remains focused and achievable.

The business goals for the website are categorized as follows:

Internal Objectives:

User-Centric Goals:

2. Scope Plane:

When considering the scope plane, we focused on what to include to satisfy all of the user needs. We applied an agile approach to building HomeIE—with a Kanban board, user stories, assigned tasks, daily stand-ups, and sprints. The wireframes and stand-ups helped create focus and develop realistic expectations. We decided on the features through extensive huddles on Slack, which provided clarity, with the tasks seeming manageable and controllable.

As per the Hackathon Guidance, we believe that the project is very scalable and thus have included details on future features that could be developed upon.

We adjusted the sprints, etc., to suit the compressed timeline of the Hackathon.

Due to time pressure, there was a danger of scope creep (adding random new features too late), so we actively avoided this by employing the MOSCOW prioritization technique; considering what items were must have, should have, and could have.

Kanban Board- MOSCOW

When distinguishing what features to add, we considered:

In terms of content requirements, we included mixed content with images, logos, video, and animations, ensuring that everything included was useful, sellable, and buildable.

Adjust to content of the end product, will we have a form, will we have video/animations/live data.

We also evaluated the project from a commercial perspective, focusing on four key aspects of business rules:

  1. Objective: What does the user want to accomplish? The user wants to find tailored advice on how to secure a home (based on their personal situation). They want to consider all angles. The platform meets this objective by leveraging data and AI to provide personalized, relevant insights, making the advice more actionable and precise.
  2. Functional: What does the user need to do to achieve that objective? The user needs to fill out a form with key details. This will be linked with the interactive AI Chatbot, which will offer tailored prompts based on the queries that the user sends to it. The user experience is designed to be intuitive and user-friendly, ensuring that users can easily navigate the process to receive the advice they need.
  3. Non-functional: What constraints does the website have? It is currently working with static data, and GPU limitations and the tight timeframe limited the size of the datasets we could train the model with. In the future, we plan to explore dynamic data integration and potential infrastructure upgrades to overcome these current limitations and enhance the system’s capabilities.
  4. Scalability: As the user base grows, can the application handle increased traffic? Yes, no PPI is stored for each user. Additionally, the application’s backend is designed to scale with the user base, utilizing cloud services and modular architecture to handle increased traffic efficiently. This ensures that as the platform grows, it can maintain performance and reliability.

User Stories Planned for HomeIE:

As a team, we brainstormed some user stories when thinking about how to build the project.

We created a shared Excel file and input user stories for consideration under the following headings:

Alt text

Sample of user stories:

User Stories Planned for HomeIE
User Story Expected Answer Format Expected Answer (if exact) Datasets
As a first-time homebuyer, I want advice on mortgage options available to me based on my income and savings. Text response with a list of mortgage options "You are eligible for the Help to Buy scheme and a first-time buyer mortgage with X Bank." RTB, Daft.ie, CSO mortgage datasets
As a user, I want to find properties in my price range in Dublin. List of properties with brief details (price, location) N/A Daft.ie, MyHome.ie property listings
As a potential buyer, I want to know the average property prices in my desired area over the past year. Text or graph showing price trends "The average property price in Dublin 8 was €350,000 over the past year." CSO, Property Price Register (PPR)
As a renter, I want to understand the potential rental yield of a property based on current market trends. Percentage or text-based explanation "The estimated rental yield for this property is 5%." Daft.ie, MyHome.ie rental listings and trends

3. Structure Plane:

The structure plane is concerned with the organization of functionality and content, ensuring that users can navigate the website effectively and intuitively.

Information Presentation

The information on "HomeIE" is presented in a structured and intuitive manner, with a clear hierarchy that guides users through the site. Content is organized logically, prioritizing the most relevant information for potential homebuyers. For example, property search results are prominently displayed, and key resources are accessible from the main navigation.

Interaction Design

The interaction design of "HomeIE" focuses on creating a seamless and engaging experience for users as they interact with the website’s features, such as forms and the AI-powered chatbot. These interactive elements are designed to be consistent, predictable, and user-friendly.

We ensured the following when designing the website:

These design choices were made to ensure that the overall user experience is not only functional but also enjoyable, encouraging users to return to the site as a trusted resource in their home-buying journey.

4. Skeleton Plane:

The skeleton plane is concerned with the navigation and interface design of the website. The primary considerations are the form that the application will take and how users will navigate through the presented content.

Key Goals:

Progressive Disclosure: We reveal information gradually, allowing users to explore content without being overwhelmed. For example, after users fill out the initial form, they gain access to more tailored advice from the chatbot.

Navigation Design: The website features a simple and intuitive navigation structure with clearly labeled tabs that guide users through the process of securing a home. For instance, users start with a form that gathers essential details, which the chatbot then uses to provide personalized advice.

In designing the website, we grouped information visually based on its importance. The form must be completed before interacting with the chatbot, ensuring that the chatbot’s responses are accurate and relevant.

We considered the Hick-Hyman Law, which suggests that offering too many options can slow decision-making. Therefore, we kept the design clear and uniform, minimizing the number of choices presented at any given time.

Good information design is often invisible—users should feel that the site is easy to navigate without noticing the effort that went into its design. We aimed to achieve this by making the website clean, minimal, and effective, ensuring that users can find what they need quickly and easily.

Use of Wireframes in Relation to the Skeleton Plane:

Wireframes were essential in planning the layout and structure of the website. Created during the research phase using Balsamiq, these 2D models provided a simplified, stripped-down version of the website. They served as a starting point, helping us decide on the form, information architecture, and arrangement of elements.

Wireframes allowed us to balance the overall design, ensure the functionality of each element, and verify that the relationships between different parts of the website made sense. They were crucial in developing the strategy and goals for the website, enabling us to refine the design based on feedback and testing.

5. Surface Plane

The surface plane is a visual language that provides information and context. It encompasses color, layout, fonts, images, order, sequence, and the overall identity of the site.

This plane is also concerned with economy—ensuring that important elements are easily recognized by users. We applied this principle by using italics and bolding to highlight key information, along with carefully selected heading sizes and structured sections to guide the user's attention.

We considered progressive disclosure, gradually revealing more detailed information as users move through the website. For example, initial property search results provide basic details, with more in-depth information available upon further interaction.

Color contrast was a key consideration. We used a purple on white color scheme throughout the website as it provides a simple yet stark contrast, ensuring that all text is legible and consistent across different devices and lighting conditions.

We selected the Roboto font after researching its legibility across various screen sizes and resolutions. This font was chosen for its clean, modern look, which aligns with HomeIE's brand identity and enhances readability, whether the text is small or large.

Visual Identity and Consistency

The visual elements of HomeIE—color, font, and layout—work together to create a cohesive identity that reflects the website's purpose and values. The color scheme is both modern and approachable, aligning with our goal of making the home-buying process as straightforward as possible. The consistent use of design elements, such as spacing and typography, ensures a seamless experience as users navigate through the site.

In summary, the surface plane of HomeIE is designed to be clean, minimal, and effective, providing users with a visually pleasing and intuitive experience that supports their journey toward securing a home.

Points of Improvement Going Forward and Scalability of the Web Application

Web Scraping and Data Gathering

Throughout the development process, we utilized web scraping to gather data from various sources. Initially, this process presented some challenges, but leveraging ChatGPT 4.0 significantly improved our efficiency in parsing and understanding the scraped data. Moving forward, we aim to refine this process further by integrating additional tools such as [mention specific tools] and automating more aspects of the scraping workflow to enhance accuracy and reduce manual effort.

Advancements in AI Capabilities

As we continue to develop HomeIE, our proficiency in utilizing AI, particularly ChatGPT, is growing. We have seen improvements in natural language processing, user intent recognition, and the overall accuracy of responses. Future enhancements will focus on deepening the AI’s contextual understanding, expanding its ability to handle more complex queries, and incorporating user feedback to continually refine the AI's performance.

Integration with Looker Studio

To provide users with real-time data insights, we plan to integrate Looker Studio into the HomeIE platform. This will enable us to create live, interactive graphs that visualize key housing market trends. These visualizations will be directly linked to the data gathered through our scraping efforts and will update dynamically as new information becomes available. Alternatively, we may directly link to the CSO API to ensure the most accurate and up-to-date data is presented.

Enhancing Chatbot Responsiveness

A key area of focus will be ensuring that the chatbot is highly responsive to the data collected through user forms. By fine-tuning the integration between form inputs and chatbot responses, we can provide users with more accurate and personalized advice. This will involve refining the algorithms that match user inputs to the appropriate AI responses and expanding the dataset used to train the chatbot to cover a broader range of scenarios.

Optimizing Information Input for ChatGPT

The effectiveness of ChatGPT in delivering valuable insights depends heavily on the quality and specificity of the input data it receives. We plan to optimize the way information is input into the system by developing more structured data entry formats, incorporating additional layers of data preprocessing, and introducing validation checks to ensure that the AI has the context it needs to provide precise and actionable advice.

Building Scalable Data Pipelines

To support the ongoing training and improvement of our AI models, we are developing a robust data pipeline system. This pipeline will allow us to seamlessly integrate new datasets as they become available, ensuring that our models remain up-to-date and scalable. By automating much of the data ingestion and processing, we can focus more on refining the AI’s performance and expanding its capabilities.

Implementing New AI Technology as it Becomes Available

We are actively monitoring advancements in AI technology through sources such as TDLR. Notably, a new iteration of ChatGPT, titled 'Strawberry,' is expected to be released soon with advanced math-solving and programming capabilities. This update will be particularly beneficial for HomeIE, as it will enhance the accuracy of budgeting and projection features, providing users with even more reliable financial advice.

Increasing Accessibility through Language

To make HomeIE more accessible, we are exploring options to include AI-driven text-to-speech functionality, allowing the website to be read aloud to users. Additionally, we aim to introduce voice command capabilities, enabling users to interact with the platform via microphone input rather than typing. We will also work on refining the chatbot’s language processing abilities to better serve users who may have limited English proficiency, ensuring that marginalized groups can fully benefit from the platform.

Strengthening the Relationship Between the Form and the Chatbot

We plan to enhance the synergy between the user form and the chatbot by introducing suggested prompts generated by the chatbot based on the information users input into the form. Additionally, the chatbot's responses to specific user questions will be tailored more closely to the answers provided in the form, ensuring that the advice given is highly personalized and relevant to the user’s situation.

Design Enhancements

As we continue to improve the visual aspects of HomeIE, we plan to introduce animations and video prompts to make the user experience more engaging and interactive. These enhancements will not only make the platform more visually appealing but also help convey complex information in a more accessible and user-friendly manner.

CREDITS

Content & Media Sources:

Acknowledgements