Free AI web copilot to create summaries, insights and extended knowledge, download it at here

2327

Abstract

er 3 seconds</b> of user input.</p><p id="2d0e">This was new. I initially assumed this meant the model should be simple. But I had not taken into account the UI and Database lags. The entire system had to respond to the user within 3 seconds of the user hitting ENTER after their query!</p><h2 id="d4f2">Product Thinking</h2><p id="dc53">Another latent “requirement” I discovered while experimenting with models was the accuracy per question. For every user query, the model may or may not always find the most appropriate answer. It generates a similarity score between each question in the database and the user’s query, and returns the top matching query.</p><p id="862e"><b>BUT</b>, not always the top matching query the most appropriate. As a Data Scientist, I could have classified this requirement under Accuracy, and would have used a complex model for better matching. But that would have led to higher response time.</p><h2 id="ec53">User’s Perspective</h2><p id="5db1">As a Chatbot user, I would be frustrated if I had to wait for more than 3 seconds after entering my query. My frustration would be equally high if I received a wrong or irrelevant response to the question.</p><p id="a8bd">But if the chatbot was confused, I would not mind if the chatbot took my inputs for clarification of my intent. This is what I called AMBIGUITY in the model.</p><p id="b871">This was neither a requirement stated by the user, nor a technical requirement.However, this requirement emerged when I put myself in the User’s shoes while testing a prototype.</p><h2 id="6cba">How I solved this Problem</h2><p id="64f9">I have mentioned the technical details of the problem in my previous posts:</p><div id="013b" class="link-block"> <a href="https://cervio.medium.com/how-i-built-the-simplest-rag-based-question-answering-system-before-chatgpt-langchain-or-b885c9dcbfdb"> <div> <div> <h2>How I built the simplest RAG based Question-Answering system before ChatGPT, LangChain or…</h2> <div><h3>A blueprint you can use!</h3></div> <div><p>cervio.medium.com</p></div> </div> <div> <div style="background-image: url(https://miro.readmedium.com/v2/resize:fit:320/1*8Ny4Gz0xnMkZeCWT8u3nAw.jpeg)"></div>

Options

   </div>
        </div>
      </a>
    </div><div id="24a0" class="link-block">
      <a href="https://towardsdatascience.com/practical-evaluation-metrics-for-a-semantic-search-bot-334f6c2f9c7e">
        <div>
          <div>
            <h2>Practical Evaluation Metrics for a Semantic Search Bot</h2>
            <div><h3>A Guide to Product Metrics in AI</h3></div>
            <div><p>towardsdatascience.com</p></div>
          </div>
          <div>
            <div style="background-image: url(https://miro.readmedium.com/v2/resize:fit:320/1*hOg7Z5tWgOyVWgCJ92yNRQ.jpeg)"></div>
          </div>
        </div>
      </a>
    </div><div id="94df" class="link-block">
      <a href="https://betterprogramming.pub/a-slightly-better-ai-based-semantic-search-bot-c806703236fd">
        <div>
          <div>
            <h2>Building a Slightly Better AI-Based Semantic Search Bot</h2>
            <div><h3>With the Roberta Sentence Tokenizer</h3></div>
            <div><p>betterprogramming.pub</p></div>
          </div>
          <div>
            <div style="background-image: url(https://miro.readmedium.com/v2/resize:fit:320/1*kqJVilNj4Sg6PImJsgCXyw.jpeg)"></div>
          </div>
        </div>
      </a>
    </div><h2 id="cc97">Lesson Learnt: UX for DS</h2><p id="22c7">Whether you are a building a Data Science Product or a Data Science solution for a Business Problem, think of it as a A highly technical, blackbox solution with which the user WILL interact.</p><p id="5ba4">Wherever interaction with a system (physical, digital or algorithmic) is involved, there needs to be a basic understanding of how the user would interact with it.</p><p id="2cb1">UX Design helps in understanding how Data products/Information based products need to be designed in order for interaction with data/information is User-friendly.</p><p id="3a52">Since I used first-principles and customer feedback on the prototype to guide me through the UX requirements, I decided to learn UX design principles so that the next time I solved DS problems, I would think of UX Design systematically.</p><p id="9313">And what better time to start than now? So I decided to pick up UX design skills in 2024!</p></article></body>

Why I am Investing Time in 2024 to Learn UX Design as a Data Scientist

Everything is about the experience! Coming from a very Data background, my focus had always been on how to make sense of data, how to hack into the power of Data Science and Data Analysis to help create value for businesses and their clients.

While trying to build a solution for a client’s problem, my focus has always been on balancing accuracy and complexity, as well as adding the additional layer of explainability.

A Data Scientist’s View of User Experience

As a Data Scientist, my understanding of user experience was purely in terms of how to make the machine’s decisions easy to understand and use for the client.

The major points of focus while designing systems were:

How to show just the right amount of data
How to build explainable models
How to educate the client on how to read and understand the output

The most logical approach to addressing the three concerns above, was to write solid documentation. After all, what user would use the system without reading the manual that came with it!

What else could User Experience mean here?

What UX Really Means in the Data World

Turns out, User Experience is a blanket term that is most easily confused with UI design, or Usage Instructions. And I had made the same mistake. I didn’t particularly pay much attention to adapting the UI design to the solution, or even spend time trying to understand how UX would apply to a Data Science Project.

I realised this only after getting a unique set of requirements from the customer.

A Fourth Dimension

I had received a request for an RAG based chatbot. As a seasoned Data Scientist, I agreed to build one for them. WE sat down to flesh out the requirements of the solution.

Instead of asking me for model explainability, or simplicity (opposite of model complexity) my client asked for a response time of under 3 seconds of user input.

This was new. I initially assumed this meant the model should be simple. But I had not taken into account the UI and Database lags. The entire system had to respond to the user within 3 seconds of the user hitting ENTER after their query!

Product Thinking

Another latent “requirement” I discovered while experimenting with models was the accuracy per question. For every user query, the model may or may not always find the most appropriate answer. It generates a similarity score between each question in the database and the user’s query, and returns the top matching query.

BUT, not always the top matching query the most appropriate. As a Data Scientist, I could have classified this requirement under Accuracy, and would have used a complex model for better matching. But that would have led to higher response time.

User’s Perspective

As a Chatbot user, I would be frustrated if I had to wait for more than 3 seconds after entering my query. My frustration would be equally high if I received a wrong or irrelevant response to the question.

But if the chatbot was confused, I would not mind if the chatbot took my inputs for clarification of my intent. This is what I called AMBIGUITY in the model.

This was neither a requirement stated by the user, nor a technical requirement.However, this requirement emerged when I put myself in the User’s shoes while testing a prototype.

How I solved this Problem

I have mentioned the technical details of the problem in my previous posts:

How I built the simplest RAG based Question-Answering system before ChatGPT, LangChain or…

A blueprint you can use!

cervio.medium.com

Practical Evaluation Metrics for a Semantic Search Bot

A Guide to Product Metrics in AI

towardsdatascience.com

Building a Slightly Better AI-Based Semantic Search Bot

With the Roberta Sentence Tokenizer

betterprogramming.pub

Lesson Learnt: UX for DS

Whether you are a building a Data Science Product or a Data Science solution for a Business Problem, think of it as a A highly technical, blackbox solution with which the user WILL interact.

Wherever interaction with a system (physical, digital or algorithmic) is involved, there needs to be a basic understanding of how the user would interact with it.

UX Design helps in understanding how Data products/Information based products need to be designed in order for interaction with data/information is User-friendly.

Since I used first-principles and customer feedback on the prototype to guide me through the UX requirements, I decided to learn UX design principles so that the next time I solved DS problems, I would think of UX Design systematically.

And what better time to start than now? So I decided to pick up UX design skills in 2024!