Good morning.

My name is Doctor Vedas.

Please start by introducing your professional journey in academia.

Nice to meet you, Dr. Vedas. When you design an introductory course in your field, what teaching method do you find most effective for helping first-year students grasp the core concepts?

What are the pract? It would include teaching is very important one. Our project based learning is very. So that the students will learn both. The theory as plus. That lab oriented. Practical. That is very important for the students. To excel in the community. In the industry possible.

You mentioned project-based learning as a key component—can you give a specific example of a project you’ve assigned that combines both theory and lab-oriented practice for your students?

Yeah, I have recently.

Thought the subject called the cloud security. That we are. Where the students will be able to do. The. The final capstone project. How the security will be incorporated in the? Thought based environment. And we used. The tool called. Palace police tool be used. For the goals, that would be Lab. And from that our students also. Developed. Yeah Mini. Broadship and submitter.

Thank you. Dr. Vedas, your PhD thesis focused on safety enhancement for distributed automotive software systems—how have you integrated findings from this research into your classroom instruction or student projects?

Then the next I have handled a subject called ethical hacking. In ethical hacking also. Most of the exercise are the project based. Learning method and the students also excel and they are submitted. 1 is a small unique project. Recently.

You mentioned your students developed a mini project and submitted it—what criteria did you use to evaluate both the theoretical understanding and practical skills demonstrated in these capstone projects?

You mentioned your students developed a mini project in ethical hacking—can you describe the process you used to ensure students adhered to ethical guidelines while conducting their practical assignments?

It is a one kind of. Said approach. That it can be used for the. Security professionals. May lead to identify. What? What is the vulnerability? Certain assistance? Before that particular malicious attackers can exploit them. So yeah. And the students? Will also be able to do this kind of uh, finding the vulnerability in different. Yeah, yes. And uh. They will do for example. One of the important thing is that the footprinting. Footprinting Sonali important thing where? Will get the information. Gathering can be done. The footcase of the footprint. And there are some different types of things like. Passive reconnaissance and the. Active reconnaissance in the footprinting. So passive reconnaissance anymore, no. Direct interactions will be there. For example, it is considered example like Google search. Or the way some social media. And in the case of the active reconnaissance, another one different, another type where there will be direct interactions will be. So for. So by one particular server or a port. So.

The story, The thing is. They will collect all the data. Like this? What up? What up the IP address? What are the? Domain details. What are the network? Stop. Apologies, all these things. And for this also there are N number of tools are available. Like who is? And Google dot in like that. So the students will be able to do their small project. And we'll submit in the. Course itself.

You described passive and active reconnaissance in student projects—how do you assess whether a student truly understands the ethical boundaries between legal information gathering and unauthorized probing during these exercises?

And rather than this footprinting, there are another method like scanning. Admirations. So the hackers. Uh, activities? What are the hacker activities are there? So the interpreting system that we can, we can easily find out the particular vulnerabilities. So for example, how many ports are open so in that way. The students be able to find. Uh, this particular type of thing called scanning one. And for this also N number of tools are able like N map. And other Nissan's like the tools are available. So like this, the students will be able to. Go in a practical manner and a project based manner. For the the particular subjects. So it will help for the students. For the placement as well As for the. Their career also.

Understood. Now, focusing on your research trajectory: over the next three years, what is your roadmap for establishing a high-impact research agenda here, especially in terms of impactful publications and building collaborations in your specialization?

Understood. Now, focusing on your research trajectory: over the next three years, what is your roadmap for establishing a high-impact research agenda here, especially in terms of impactful publications and collaborations within your specialization?

Actually, regarding the research.

As of now I'm having in corpus. 30 papers.

Let's shift to teaching approaches. Imagine you are assigned a large introductory class in your specialization, but you cannot use traditional lectures or slides. Describe in detail how you would implement an active learning model, such as a flipped classroom, to maximize student engagement and understanding.

Thank you for sharing that, Dr. Vedas. Now, let's discuss institutional stewardship. Suppose the department needs to improve its research publications per faculty metric to meet international accreditation standards. What systematic approach would you implement to help achieve this goal?

Corresponding as well as the first author. And the ADAB 66 papers as as uh corresponding on them. And my plan over the. Next three years about the researchers have been but. To to give submit a proposal. In the government search bodies like. PS Team and ICMR and other things. And to get the fund from them. And already I have submitted some 3 research. Uh, proposals? It is on the way. The middle of processes past brief. First level and this is going where I'm waiting for the. A particular interview or presentation based. So my thing is to publish.

High impact factor. Research papers in my domain. My domain is a cyber security domain. And the cyber physical system. Domain and. And this is the.

You mentioned submitting research proposals to funding bodies like DST and ICMR—can you walk me through the documentation or processes you use to ensure your proposals meet all compliance and accreditation standards required by these agencies?

You mentioned aiming to publish high-impact research in cyber security and cyber physical systems—what strategies will you use to build collaborations with other departments or industry partners to strengthen both publication quality and accreditation outcomes?

Yeah, actually cybersecurity is not.

A single domain. Like. It is a. We can have that cybersecurity domain in some other.

Thank you for sharing your approach. Now, Professor, imagine you are faced with a situation where you are accused of grading bias by a student, but your department is under pressure to maintain a certain pass percentage for accreditation. How would you handle this conflict between upholding academic integrity and meeting institutional expectations?

Right, so with cybersecurity being multidisciplinary, can you give a concrete example of how you would initiate a joint research or student project with another department—say, collaborating with electrical engineering or management—to address a real-world problem?

Interdisciplinary. Interdisciplinary environment. RY7 for cyber physical system is the one environment. Where it is interdisciplinary, not. Specific to one particular domain.

Right, so with cybersecurity being multidisciplinary, can you give a concrete example of how you would initiate a joint research or student project with another department—say, collaborating with electrical engineering or management—to address a real-world cyber physical system challenge?

So other departments. Faculties. Or even industry persons. They can contribute their. There were portions in the. Publications are. One of the rest of the case of the. Audit proposal also that. Particular project will be a high quality project whenever the. Other peoples are collaborated. Other industry person is involved in the. So it will be a collaborative 1 so. That is the thing.

Moreover, my current. Proposal also we have. Signed on in the in our Bangalore.

That makes sense. When you involve industry partners in these collaborations, what specific criteria do you use to select which companies or professionals to work with?

So, so that particular company. They will also enroll in my project. When the uh, when it is being developed. So the particular project will become. Outcome will be. High impact.

11 company here.

Understood. In your current proposal involving a Bangalore-based collaborator, what mechanisms have you put in place to ensure clear authorship roles and resolve any potential conflicts of interest during publication or project execution?

Thank you. Given your experience guiding students through hands-on cybersecurity projects, how do you ensure the evaluation process remains fair and unbiased, especially when assessing projects that may differ significantly in technical depth or approach?

Thank you for clarifying your approach. In your current industry collaboration, what steps do you take to ensure that student project outcomes are both academically rigorous and directly relevant to the partner company’s needs?

Actually student projects.

Understood. When assessing these student projects, especially those involving real company data or proprietary tools, how do you address potential conflicts between institutional grading standards and specific outcomes the industry partner wants to see?

Constant so in that. Manner what are the their particular? Problems they have specified. With the students level. And we will guide so our students to. Uh, do the work and that. Even the like that also we have. I think it's some two years before also we have.

Let's shift focus. Professor, what concrete strategy would you use to leverage your professional network to create new internship or employment pipelines for students, ensuring their skills align with current industry needs?

And they got up around 8000. As a sponsor from the.

Let's discuss your expertise in multimedia or artificial intelligence as applied to media. Can you describe a specific project or research contribution where you applied advanced multimedia or AI techniques, highlighting your role and the impact on the field?

Uh, nothing but.

Understood. Let’s talk about your experience teaching theory and laboratory courses—can you describe how you structure a new course to balance foundational concepts with practical application, and give an example of how you’ve implemented this in the classroom?

Predict the outcomes. And they can also reduce the. Human efforts also. So this can be applied for any kind of. The software projects. Or even in the case of the business projects also we can use this part. I think.

Understood. When integrating AI tools to automate decisions in your student or business projects, what criteria do you use to evaluate whether the AI’s recommendations are sufficiently reliable to be trusted over manual decision-making?

And the most moreover in the AA related projects. We'll use the machine learning. Concert. So that there will be some kind of. Atkins and the UH. Add data. And they can make some kind of predictions. For example. The ML there are a number of. Types are supervised learning. Unsupervised loving. Reinforcement. Learning. So in the case of the supervised learning is nothing but classifications, progressions can be done. And in the case of the unsupervised learning clustering. So like that, there are a number of different types of things. So we can predict.

One particular projects. What about the risk? And delayed by this particular ML related things. And even the by using this ML algorithms we can estimate the cost and the time. And what are the customer behavior in this client so?

You mentioned using machine learning techniques like supervised and unsupervised learning for prediction—can you give a concrete example of a project where you selected the specific ML approach, and explain what data you worked with and how it improved the outcome?

These a tools and these things we we can. Predicting can be done. The weather in one particular software project. Will minimize their deadlines to the. Like the?

Okay, you mentioned estimating cost and time using machine learning algorithms—what specific features or data points did you find most predictive for project delays in your work?

So another thing is. Uh yeah, tools we can use, even the NLP natural language processing can also be done. Where we can understand the?

Understood. When you apply machine learning for project risk or cost estimation, how do you validate that the predictions are accurate enough to inform real decisions?

The process of human languages. For example, let us see the chat port.

Right, you mentioned using NLP for understanding text—can you share a concrete example where you applied NLP techniques to a real dataset, and describe the outcome or insight you were able to generate?

You mentioned chatbots as an example of NLP—what specific challenges did you face in handling ambiguous or multi-intent user queries, and how did you address them in your implementation?

What's the process? Of the human language.

Understood. Earlier you mentioned using machine learning to predict project risks and delays—can you walk me through how you selected which features or data points were most important for training your model?

And for the customer support purpose. Chat bot. Is a very useful. That is our one important projects. And even for documentation purpose.

Understood. When you implemented the chatbot for customer support, what was your approach to evaluating its effectiveness in real interactions with users?

Requirement analysis be done from the. Government patients. And sentimental analysis. Of user feedbacks so these are something we'll. In the process, the main thing is. Extracting. Some kind of requirements automatically?

From the emails also that can be done by the help of this.

Extracting requirements automatically from user feedback sounds valuable—what specific techniques or tools did you use to perform sentiment analysis or requirement extraction, and how did you ensure the results were accurate?

Understood. For the requirement extraction from emails that you mentioned, what approach did you use to differentiate between explicit requirements and more implicit or ambiguous feedback in the text?

Requirement Requirement extraduction.

Got it. Let's shift to your experience teaching theory and laboratory courses—can you walk me through how you structure a typical lab session to ensure students not only complete the tasks but also deeply understand the underlying concepts?

Requirement Requirement extraduction.

Got it. When extracting requirements from emails, how did you validate that the automatically identified requirements matched what stakeholders actually intended?

Correct. Actually, in the case of requirement extradition. From emails. It's nothing, but it's a one kind of process.

Understood. When you automated requirement extraction from emails, how did you handle situations where stakeholders used domain-specific jargon or abbreviations that the system might not recognize?

The process of requirement. The extractions normally will done by this. Collecting of emails. I'm gathering emails from the stakeholders. And filter the relevant threats, we can do it. Then the next stage is limit processing. When the cleaning. So where? Will be removing the signature. I will get some kind of. Uh, spams. And other things and in the case of the cleaning also preprocessing also will do. Some kind of tokenizations? For example. Printing the text as well as into the words. Or stop and stop a word removals. Like something we'll be doing. So all this. Text, uh, ready. Uh, making a picture leave for one kind of analysis purpose. Is the inventory, then becomes the. After pre processing information. Extraction with the help of the NLP Natural animal Processing.

You described preprocessing steps like tokenization and stop word removal for requirement extraction—can you give a specific example of how your approach handled a particularly messy or unstructured email thread and what structured output it produced?

Techniques and 1st we'll give the keyword extraction can be done. What's the login? What's the payment? About the security. Then comes the named entity organizations. Where the data updates names as well as your models to be there. Then comes the techniques like intern deductions. Some complaints request. And the future. So all this. Identifies.

Some actual requirements hidden in the. Sentence. So these are the main process happening. It is in the developer stage. During my project it's not a complete state. These are. One kind of developers. Stage process.

You described using techniques like keyword and named entity extraction for requirement gathering—how did you evaluate the accuracy of your extracted requirements against the actual project outcomes?

Understood. You mentioned this process is still in the developer stage—can you describe a specific challenge you've encountered during development, such as handling overlapping requirements from multiple stakeholders, and how you're approaching it?

Challenges in the sense. I'm so flow.

You mentioned being in the development stage—can you give a concrete example of how you handled emails containing conflicting requirements from different stakeholders during your extraction process?

Some ambiguous. Ambiguous languages in emails will be that what is the one challenges you face? Then.

Some contexts. Is missing. Across. Different threads.

Ambiguous language can definitely pose a problem. Can you walk me through a real example where ambiguous wording in an email led your extraction process to misinterpret a requirement, and how you discovered or corrected the issue?

Right, dealing with ambiguous language and missing context is tricky—can you give a concrete example of how you tried to reconstruct the missing context from scattered email threads during requirement extraction?

Is another one challenges here? Then.

Some datas. Require some more training.

Right, missing context across threads is a real issue—can you describe a practical method or specific strategy you used to reconstruct or infer that missing context so your requirement extraction would be more accurate?

Is that some few challenges? Play Makan sofa.

Thanks for outlining those challenges. You mentioned that some data requires more training—can you explain how you determine when your current dataset is insufficient, and what steps you take to improve your model’s performance given those limitations?

Understood. When you realize your current data set needs more training, what criteria do you use to decide which new data samples to collect or prioritize first for model improvement?

In in that way within I didn't go for so far.

I will. I will. Take into considerations.

Alright. You mentioned the process is still in development—how are you ensuring that the extracted requirements are categorized correctly as functional or non-functional during this stage?

Alright. Considering you’re still early in the process, how are you validating the accuracy of the requirements you’ve extracted so far—are you using any manual review or feedback from stakeholders at this stage?

Actually this. I couldn't see UH will be done by the feedback from the UH stakeholders. And.

By the help of the thing like.

Thanks for clarifying. Since stakeholder feedback is your main validation, what have you observed as the most common type of correction or suggestion they provide on your extracted requirements?

Understood. Earlier you mentioned using NLP techniques like keyword extraction and named entity recognition—can you give a concrete example of how you structured a particularly ambiguous requirement after running it through your current pipeline?

Actually in the my thing e-mail.

The system should allow.

Login.

Thanks for clarifying. Earlier, you mentioned that email threads sometimes lack context—can you describe how you address this issue to ensure your NLP models can accurately extract requirements despite incomplete information?

Could you walk me through how you take an ambiguous email instruction like “the system should allow” and refine it into a concrete requirement using your NLP pipeline?

And respond within 2. Seconds.

Got it. When you encounter a requirement like “the system should allow login” in an ambiguous email, what steps do you take to clarify and document the specific functional details needed for your development team?

That is the one small thing. And for this purpose. Let's take, yeah. Functional thing. Login features non functional thing as their response time.

Okay. Given your requirement for the system to allow login and respond within 2 seconds, how do you specify or enforce this performance constraint during both development and testing phases?

You identified response time as a non-functional requirement—can you describe a method you’ve used to measure or monitor this login response time in a real deployment?

Artificial intelligence output. To extract. Both functional, non functional correctly.

That is nothing but that accurate. Accurate is very important.

That distinction between functional and non-functional requirements is important. Can you describe how you would communicate these extracted requirements clearly to a software development team to ensure they’re implemented accurately?

So for example, any missed performance requirements are missed. Automatically it will become a lower recall. Or any added requirements or wrong means. It will become a low low positions. So that is why. This should be like a status should be. Accurate. For both function. And non functional 1.

Right, accuracy is crucial for functional and non-functional extraction—when your system misclassifies a requirement, what’s your process for correcting it and retraining your extraction model?

And some factors which may affect. This, like you crazy, is nothing but.

You emphasized the importance of accuracy and recall in requirement extraction—can you describe a time when your extraction system’s low recall or precision directly impacted a project outcome, and what you did to address it?

You mentioned that factors like accuracy and recall can affect your extraction system—can you give an example of a specific metric you use to quantify these, and how you interpret changes in that metric during evaluation?

So all the that is important. Then the next thing is something but natural language. Processing model. Which model is being used?

Example any? Basic keyboard matching is there. That will become a lower accuracy. Any advanced model? Is used for example that is cancel a transformer.

Right, the choice of natural language processing model matters a lot—can you walk me through how you evaluated and selected the model you’re currently using for requirement extraction?

Transformer like. Perth or GPP means automatically it may give a higher accuracy.

You pointed out that using just keyword matching can lower accuracy compared to advanced models like transformers—can you describe a situation where switching to a more complex NLP model directly improved either recall or precision for requirement extraction in your project?

So that is the. I think your model use in every model mainly used, so we have to be. Based upon that particular requirements.

You mentioned that the quality of email data can affect extraction accuracy—can you give an example of a specific data quality issue you encountered, and how it impacted the model's performance?

Yeah, that's something.

You mentioned that choosing the right NLP model is critical for requirement extraction accuracy—what process do you follow to validate that a new model, like a transformer-based one, is actually improving extraction accuracy on your real email data?

You’ve emphasized the importance of choosing the right NLP model and mentioned that transformer-based models can significantly improve accuracy—can you share a specific example where upgrading your model led to a measurable improvement in extracting either functional or non-functional requirements from emails?

Yeah, as I told that functional requirements.

Right, so in your experience, what concrete steps did you take to evaluate the model’s improved extraction—did you run a side-by-side comparison with previous outputs on a shared test set, or use another method?

They run side by side comparisons.

Understood. During your side-by-side comparison, what specific metrics or criteria did you use to determine whether the new model was more accurate at extracting requirements?

Actually in the side by side comparison. Model.

Okay. When you compared the models side by side, how did you ensure the test set of emails represented a realistic mix of requirement types and data quality issues?

The first matrix. Precision.

So precision. Is something but. The correctness of the extracted requirements. So there is a one kind of matrix.

Precision makes sense as a metric—how did you handle cases where two models disagreed on whether something was a functional or non-functional requirement?

Which we? Came across. Then comes the recall. Metrics. Sorry, call matrix something but. Completeness of the extractions. Is the recall matrix.

Precision makes sense as a metric—how did you handle cases where two models disagreed on whether something was a functional or non-functional requirement?

Done your fun scope. So F1 score is another one matrix. We loosened our project. Where to balance the precision and the recall?

Okay. Earlier you mentioned the importance of data quality in email inputs—can you describe a situation where poor data quality directly impacted your model’s ability to extract requirements, and how you addressed it?

Got it. When you calculated the F1 score for your model’s performance, did you notice any particular scenarios or types of requirements where either precision or recall consistently lagged behind, and how did you address that?

And I couldn't see the overall correctness is nothing but accuracy.

And these are the metrics? Which we came across in our project. For the e-mail instruction.

You mentioned accuracy as an overall metric—when interpreting the results, how did you decide whether to prioritize accuracy, precision, or recall for your requirements extraction use case?

Understood. When you were evaluating the model on these metrics for email extraction, did you encounter any challenges in labeling the ground truth requirements, especially given the potentially ambiguous language found in emails?

Yeah, actually during our. This e-mail extraction.

Thank you for clarifying. When you encountered ambiguous or unclear requirements in emails, what process did you use to resolve disagreements among team members about the correct label?

Most of the emails. Are rarely written in the formal. Requirements.

What is the one thing even maybe in a structured unstructured data? Because emails. Most probably will contain some signature. Some greetings will be there. Some other. Things will be the replies like they'll be there so. All information is. Scattered across multiple threads. So that unstructured data is another one challenges. Yeah. And coming to the context. A loss.

Given that most emails weren’t written in formal requirements language, can you describe a specific technique you used to standardize or interpret ambiguous phrases during the extraction process?

Right, considering the unstructured nature of email threads, what specific preprocessing steps did you use to filter out signatures, greetings, or irrelevant content before passing the data to your extraction model?

Then across the different threads. For example one e-mail ID. E-mail have. Some login features. And another e-mail, uh. Uh, we'll have. So using some OTP. Or an e-mail may have only that particular mobile user, so. These are the some. Few challenges be. Came across.

You’ve highlighted that unstructured content like signatures, greetings, and scattered replies adds complexity—can you walk me through a specific method or tool you used to reconstruct the full context of a requirement spread over multiple email threads?

It may be with respect to the finance means finance. Healthcare Cloud. What other baby?

The upper depends upon the particular one.

Okay. In your experience, when dealing with requirements scattered across domains like finance or healthcare, did you use any automated tools or was the contextual linking done entirely manually?

Okay. When you needed to connect related requirement fragments from different emails within the same domain, did you rely on any natural language processing techniques or was it mostly based on manual analysis?

MMM, so in the requirements. Fragments from different uh.

MMM, so in the requirements. Fragments from different uh.

Right. Earlier you mentioned context loss across threads—did you ever use any clustering or thread-reconstruction algorithms to help group related emails together during your extraction process?

Understood. When you were piecing together those requirement fragments, did you ever use techniques like entity recognition or topic modeling to assist in linking related content, or was it all manual identification?

Yeah, yeah, yeah, actually.

Right. When you were reconstructing these requirements from fragmented emails, did you ever encounter conflicting information, and if so, how did you resolve those conflicts to determine the correct requirement?

Yeah, actually the requirement fragments. Uh, different places, For example one e-mail maybe?

Got it. When you gathered requirement fragments from various emails, what approach did you use to ensure that nothing important was missed when consolidating them into a single set of requirements?

Yeah, so, uh. So. Uh, another, maybe some more mobile related or, uh. Things so whenever the real. Time projects.

You mentioned real-time projects where requirements might be split across emails for things like user logins and OTPs—can you describe a concrete example where you had to synthesize such fragmented information into a clear technical specification, and what your first step was in approaching that task?

So.

NLP role is a very important role in identifying these kind of requirement. Fragmentations because NLP it will detect. The requirements. Related What are the sentences? Are there and they'll extract the particular entities. Then they will link the all the. Uh, fragments. Uh, across the emails.

You mentioned stakeholders often write requirements in a gradual, informal way—can you give a specific example of how you clarified or formalized an ambiguous or incomplete requirement directly with a stakeholder?

And they'll resolve this kind of.

That makes sense. When using NLP to extract and link requirement fragments, which specific entity recognition models or techniques did you find most effective for handling informal or fragmented stakeholder emails?

Right, so when applying NLP to extract requirements from informal emails, what challenges did you face with entity recognition specifically, and how did you address those challenges?

And moreover, the NLP techniques normally we'll be using, the tokenization is the NLP technique we'll be using.

You mentioned tokenization as a key NLP technique—can you share a situation where tokenization alone was insufficient to extract requirements accurately, and what additional method you applied to overcome that limitation?

As we already know, tokenization breaks a particular text into. Some words and sentences. Norwegian may be a named entity organization also we can do. Or dependencies parsing. Can also be.

Right, so when you applied dependency parsing to these fragmented requirements, can you give a concrete example of how it helped you reconstruct the intended workflow or identify missing components in the process?

NLP techniques. Being used from the e-mail extraction.

You’ve mentioned using named entity recognition and dependency parsing to extract requirements from informal emails—can you provide a specific example where dependency parsing revealed a relationship between requirement fragments that simple entity recognition would have missed?

Yes, in this particular process. Of requirement.

Fragmentation fragment. It's not. It's not action. The first step. We have to collect all the emails.

Right. When you used these NLP techniques on real project emails, how did you validate that the extracted requirements were both complete and accurate before progressing to implementation?

Right, so after collecting all the emails, what process or tool do you use to ensure that none of the informal requirement fragments are missed before moving on to NLP-based extraction?

Gather the related emails. Tricks.

The second step is a being bad. You have to preprocess. The particular text. For example. If any greetings that particular greetings. Any signatures that we have to remove the particular signatures? Any other four quarter? Complex is that. The third step I think but. We have to detect. What is the candidates? Fragments, all that. Again, The thing is, you're defying some kind of sentences. It's being likely to contain one particular requirements information, so that's in the third step.

Understood. Earlier, you mentioned that stakeholders often provide requirements in a gradual and informal manner—can you walk me through how you ensure that no key requirements are overlooked during this initial collection phase, especially when fragments are scattered across multiple emails?

Then the 4th step is having but we have to extract the attributes.

Okay. For the preprocessing phase where you remove signatures and greetings, what specific rules or techniques have you implemented to reliably distinguish between actual requirement content and these non-informative parts of the emails?

Understood. For your fourth step of extracting attributes, what specific challenges have you faced when trying to identify and link attributes that are implied rather than explicitly mentioned across multiple fragmented emails?