Hemali Dodia

Could you help me to understand more about your background by giving a brief introduction on yourself? Okay, so my name is Himali Doria, and I have worked on multiple various projects, including data analysis, data science, and automation. Currently, I'm also working on a data analysis project where I'm helping a client decide who to give a loan to and whom not to give based on risk analysis, which involves determining the risk and the type of loans they can disburse. Apart from that, I also do web scraping from various websites. I scrape data, clean it, and put it in a proper required format, then pass it on to the client or end customer. In addition, I have developed chatbots based on LLM and lantern models, similar to chatbotix AI. And, I have achieved an accuracy of up to 80 to 90%.

I prefer to monitor all the time durations by which the bot is giving the answer because when I create my own chat, like, my chatbot for my client, so at that time, it was taking a lot of time, approximately 20 to 28 seconds, which is not good. So, actually, it's not good at all. I prefer to monitor whether the bot's answer is accurate apart from that, how speedy, like how quickly it is giving an answer to the customer, whoever is chatting with the bot. And to optimize the performance, I will first work on its speed and accuracy, because these are the two parameters that matter the most.

To optimize an SQL query that aggregates data across multiple tables, I will try to avoid using more than one database. Many scenarios require data from different databases or sources, but I will try to minimize the number of joins by creating views. If I have to pull data from different databases or sources, I will create a view for that particular database, write one query, and fetch answers from the view table rather than the actual database. This will optimize my answers. Suppose my query is taking a lot of time, 2-3 minutes. I will try to reduce or avoid multiple joins and databases. I will write simple queries like select * or select multiple columns from a single table with specific conditions. I will prefer keeping all queries simple and optimized rather than applying multiple joins.

How would you refactor a chatbot score base to measure the solid principles? I would like to prefer a first entire go through the entire core, and then I would try to reduce core at most as much as possible. I can just find out the most generic or common thing which is occurring repetitively. And based on those things, it will optimize my time as well as space complexity, and it will adhere to the solid principles. I will try to reuse code as much as possible in most optimized way. I would try to reduce if there are lots of loops and lots of conditions inside it. Let's say, a nested condition within a nested loop, or more than two nested loops. I would try to minimize those things. I would like to implement more optimized functions that are currently available up to the date.

Name of design pattern that would be suitable for a real time chatbot. Message handling and deeply explain why. Okay. So as per me, when it's a real time chatbot, this for an example, just let chat, Jupyter, or something like that, then I would prefer the pattern like whenever as soon as I get any message thread, depending on that thread, I would like to get an answer. If suppose there are forming questions, there are most frequent question or generic questions, I would not like to, you know, get it from the database. There is it is if it is very handy, it will give response very quickly. And if I have a good server and everything, then I would like to, you know, student in a this one format or some Excel or somewhere. For the most generic questions, which will be very few. If there are no if there are lots of questions, then I would like to prefer the most quickest database possible. And I would like to prefer that as soon as I get any question, I remove all the punctuation. I correct the grammar so that my bot can understand it very easily. And, I provide the most suitable or the accurate prompt for my bot to answer them. For my current, you know, chatbot, I used prompt engineering. And the more accurate my prompt is, the more better answer I used to get in more quick manner. So I would like to, you know, work on more prompting for my bot.

Okay, I would like to use the classic classification method because slang and non-standard languages require a lot of classification, whether it is slang or not, and what exactly it means. For those scenarios, I would prefer a classic classification approach with proper labeling that this one is slang, which means so and so thing. Apart from this, I would like to use prompting as well because understanding these lines will require some description to the bot. So, those descriptions I can take from my database. But, as per the NLP model, I would like to use classification more often. And, I would like to use RNN for this task because I guess RNN will be the best suitable method or NLP model to understand text data.

At the service, a chatbot is used to identify and explain the mistaken house. Okay. I'm not so aware about the asynchronous handling because this is completely in JavaScript. But if I'm not wrong, an async function should be there. So, there is a function, getting user input, then setting a timeout. Okay, that is fine. Writing an event listener for user input, for example, dot value. I'm not so sure.

A Java function for NLP has a logical bug. Identify the mistaken menu in the streaming process. Okay, yeah. I just ask myself, there is some issue with the stem, stems don't add brackets, words dot substring from 0 to word dot length of length minus 3. There is some logical error.

When scaling an AI chatbot for handling millions of users, the critical aspect to consider is the hardware where we are implementing these blocks because the systems that we are using are not sufficient enough. They can easily get crashed. While I was working on a chatbot, the system was not up to date or up to the mark required to handle millions of users. It used to crash, it used to hang, and it made us lose code in multiple things. So, system requirements should be a top priority. Apart from that, I would prefer that all systems have proper security measures, such as IT security, to prevent any threats. Next, my database is crucial. I would prefer that all databases are in a highly structured format because the more structured the data, the quicker the answer I get, and it reduces user wait time. Users never like to wait, so this is a critical aspect that my chatbot should maintain while scaling. It should not compromise on speed and quality that it provides. If it compromises on either, user experience will be affected, and it will badly impact the company. So, I would prefer that it maintains accuracy and speed, either improving it or maintaining the same, but not decreasing it.

Okay. So for voice recognition, there are medical libraries that I would like to use. For an example, Librosa, Weka. I would prefer to do it in Python because I'm good with that. And I have already worked on voice recognition and voice understanding, speech to text, and text to speech processing. So I would like to prefer them. While processing, there will be noise. So first, I would remove all the noise. Then, I would perform sampling and other cleansing processes on the data. After all the cleansing processes, I would perform the necessary tasks depending on the problem statement. Depending on the requirement, as soon as I clean my data, I would proceed. It could be to understand a particular record, predict something from the voices, or determine something.

An understanding of a graph database would benefit the development of the AI chatbot. Majorly, graphical representations are best for any understanding rather than just normal text data or just communication because graphs allow us to view it from different perspectives. It allows us to see from the client perspective, from the company perspective, from the end-user perspective, everything. So, with a graph database, we can see how many users don't like our stuff or have complaints, who likes our features, which feature they like most, which feature they hate most, where they get stuck, improve, where our costs are getting affected, where our time is being spent, what is taking a lot of time, what is giving inaccurate answers. A graph database helps a lot.