Professional Bio
This portfolio is presented by Krishna Saketh and brings together selected coursework artifacts that demonstrate learning, experimentation, and practical application.
I am a graduate student at Indiana Wesleyan University developing my knowledge and skills in artificial intelligence and machine learning.
Through my coursework, I am building an understanding of how AI and machine-learning methods work, where they can be applied, and how to communicate technical ideas clearly. I am interested in connecting concepts to practical examples and understanding how these technologies can support real-world problem-solving.
This portfolio will grow throughout my program as I add coursework artifacts, reflect on what I learn, and demonstrate my progress.
Personal Value Proposition
I aim to make AI and machine-learning concepts understandable and useful by connecting technical ideas with practical problems.
I focus on breaking down complex topics into clear explanations and using familiar examples to show how AI/ML concepts work.
This approach supports communication with technical and non-technical audiences and helps connect machine-learning concepts to practical applications.
Coursework Artifacts
This portfolio presents selected coursework demonstrating foundational understanding of AI, machine learning, deep learning, and practical AI tool development.
AI Lab: Building ThinkBot, a stage-based design thinking chatbot using generative AI tools and a rule-based prototype.
View Artifact 1Machine Learning and Deep Learning: types of machine learning and a comparison of ML and DL approaches.
View Artifact 2Deep Learning and Neural Networks: understanding how a neural network learns to estimate house prices.
View Artifact 3Artifact 1: AI Lab — Building a Design Thinking AI Chatbot
A course project exploring generative AI tools, research assistants, and the design thinking process to plan, build, and test ThinkBot, a stage-based chatbot for students learning design thinking.
AI LabGenerative AIPrompt EngineeringDesign ThinkingHTML/CSS/JavaScriptIntroduction
This artifact documents an AI Lab activity focused on using generative AI tools and a human-centered design process to create ThinkBot. The work included comparing responses from a large language model, exploring a research-focused Custom GPT, using STORM AI to investigate a topic, and applying the five stages of design thinking to a chatbot prototype.
Description
ThinkBot is a browser-based chatbot prototype intended to help students and early-career professionals work through the five stages of design thinking: Empathize, Define, Ideate, Prototype, and Test. The project began with experimentation and research, moved through problem definition and idea selection, and resulted in a working prototype that was tested by three classmates.
The complete original AI Lab documentation is available here: Open the full AI Lab submission.
Objective
The objective was to investigate how AI tools can support learning and problem-solving, then apply the design thinking framework to create a practical chatbot that provides real-time, stage-specific guidance. The intended users were students learning design thinking and early-career professionals who may be unfamiliar with its terminology and sequence.
Process
- Experiment with a large language model: ChatGPT was given a prompt structured around Context, Task, and Constraints. Three separate outputs were compared for tone, depth, usefulness, and completeness. The comparison showed that the same prompt can produce different responses; the most useful ideas included practical examples, How Might We (HMW), SCAMPER, and the iterative nature of design thinking.
- Explore a Custom GPT: Consensus was used to explore research on AI-assisted design thinking. The interaction included narrowing the topic, adding a personal viewpoint, requesting more recent studies, asking why a study was selected, and considering whether findings would apply in a corporate setting.
- Use an AI research assistant: STORM AI was used to research design thinking in AI product development. Its generated article and research outline were reviewed, and the work noted the need to verify citations and check for recent developments.
- Empathize and define: The target users, their pain points, and ethical considerations were identified. The central problem was that students need real-time, stage-specific guidance because static course materials cannot respond to the particular problem they are trying to solve.
- Ideate and select a solution: Several concepts were considered, including a quiz, document-based assistant, visual canvas, collaboration tool, role-play chatbot, and stage-based chatbot. The stage-based conversational chatbot was selected because it addressed the core need and was feasible to build within the assignment scope.
- Build the prototype: ThinkBot was developed as a standalone HTML, CSS, and JavaScript web page and hosted on Netlify. It uses keyword detection and a rule-based response library of more than 20 curated responses, rather than a live LLM backend. Its interface includes stage navigation, suggested prompts, a welcome card, and chat-style messages.
- Test and review: Three classmates tested the chatbot. Feedback supported the clarity of the stage buttons and usefulness of suggestion chips. Testers also identified a mobile scrolling issue and requested conversation memory. These were recorded as areas for further improvement; conversation memory remains a limitation of the current rule-based prototype.
Tools and Technologies Used
Compared outputs from three prompt sessions and explored prompt design.
Explored research about AI-assisted design thinking through an interactive research conversation.
Generated a structured research article and outline on design thinking in AI product development.
Implemented the chatbot interface, curated response library, and keyword-based interaction logic.
Hosted the standalone chatbot and made the prototype accessible through a public URL.
Implementation note: ThinkBot is a rule-based prototype and does not make external API calls. It simulates chatbot guidance through keyword detection and predefined responses. It should not be described as a trained deep-learning model or a production LLM application.
Value Proposition
Unique Value: This project brings together prompt experimentation, AI-assisted research, human-centered problem definition, and a working web prototype. It demonstrates the ability to evaluate different AI tools rather than accepting their outputs without review, and to translate user needs into a practical, testable solution.
Relevance: The artifact is relevant to AI/ML because it demonstrates responsible use and evaluation of generative AI tools, an understanding of the limits of AI-generated research, and the application of structured problem-solving to an AI-related product concept. It also demonstrates basic web-prototyping skills and user-feedback-driven iteration.
Testing Outcomes and Further Improvements
- Classmates found the stage navigation clear and the suggestion chips helpful when starting a conversation.
- One tester reported that suggestion chips require horizontal scrolling on a phone; a clearer scroll indicator was identified as a possible improvement.
- A request for conversation memory highlighted a limitation of the current rule-based design. A future version could explore an LLM API and context management, with appropriate privacy safeguards.
- The project emphasizes that AI guidance should not replace real user interviews, human feedback, or professional facilitation.
References and External Evidence
The following sources provide external support for the design-thinking process, prompt iteration, and responsible AI considerations discussed in this artifact. They are supporting references; they do not independently validate the ThinkBot prototype or the original research outputs.
- Stanford d.school. (n.d.). Design thinking bootleg. https://dschool.stanford.edu/tools/design-thinking-bootleg. Supports the Empathize, Define, Ideate, Prototype, and Test process used to structure the ThinkBot project.
- OpenAI. (n.d.). Prompt engineering. https://developers.openai.com/api/docs/guides/prompt-engineering. Provides guidance on clear instructions, structured prompts, examples, and iterative prompt refinement relevant to the LLM experiments.
- Tabassi, E. (2023). Artificial intelligence risk management framework (AI RMF 1.0) (NIST AI 100-1). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.AI.100-1. Provides a recognized framework for considering trustworthiness and risks in AI systems, relevant to the prototype’s scope and safety boundaries.
Source note: The original submission discusses ChatGPT, Consensus, and STORM AI outputs. Any specific studies or statistics mentioned in those generated outputs should be individually checked against their original publications before being treated as evidence.
Artifact 2: Types of Machine Learning and Machine Learning vs. Deep Learning
An overview of the main machine-learning learning approaches, followed by a comparison of traditional machine learning and deep learning through house-price prediction and machine translation.
Machine LearningSupervised LearningUnsupervised LearningReinforcement LearningDeep LearningIntroduction
Machine learning (ML) enables computers to learn patterns from data and use those patterns to make predictions or decisions. Different machine-learning approaches are selected depending on the type of data available and whether the desired outcome is already known. Deep learning (DL) is a subset of machine learning that uses neural networks with many layers to learn useful representations from data.
Description
This artifact has two parts. The first introduces four common types of machine learning. The second is based on the submitted report, Machine Learning vs. Deep Learning: Choosing the Right Approach, which compares a structured-data prediction task with a complex language task.
Part 1: Types of Machine Learning
1. Supervised Learning
Supervised learning uses labelled training data, where each example includes input information and a known correct output. The model learns the relationship between inputs and outputs and then predicts outcomes for new data.
Examples: predicting house prices from property details (regression) and classifying an email as spam or not spam (classification).
2. Unsupervised Learning
Unsupervised learning works with data that does not have labelled answers. The model looks for patterns, similarities, groups, or unusual observations within the data.
Examples: grouping customers by purchasing behaviour (clustering) and identifying unusual transactions (anomaly detection).
3. Semi-Supervised Learning
Semi-supervised learning combines a smaller amount of labelled data with a larger amount of unlabelled data. It is useful when labelling every example would require significant time, effort, or expert input.
Examples: classifying images when only some images have labels, or categorising documents when only a small portion has been manually tagged.
4. Reinforcement Learning
Reinforcement learning trains an agent to make decisions by interacting with an environment. The agent receives rewards or penalties based on its actions and learns a strategy that aims to maximize cumulative reward over time.
Examples: an agent learning to play a game, a robot learning how to navigate, or a system learning a control strategy through repeated interaction.
Summary of the Four Types
| Type | Training information | Main purpose | Example |
|---|---|---|---|
| Supervised | Labelled data | Predict a known target | House-price prediction |
| Unsupervised | Unlabelled data | Discover patterns or groups | Customer segmentation |
| Semi-supervised | Some labelled and much unlabelled data | Learn with limited labels | Image classification |
| Reinforcement | Rewards and penalties from interaction | Learn actions or a strategy | Game-playing agent |
Part 2: Machine Learning vs. Deep Learning
The comparison below follows the supplied report. Traditional machine learning often relies on people to select useful features and can work effectively with structured data and moderate datasets. Deep learning uses multi-layer neural networks to learn features from data and is particularly useful for complex, unstructured inputs, although it generally requires more data and computing power.
Example 1: Machine Learning — House-Price Prediction
Regression-based models can estimate home values using structured property information such as square footage, bedrooms, bathrooms, lot size, age, and neighbourhood. Linear regression is suitable when the important features are already understood, the data is tabular, and the result needs to be explainable. A deep neural network may add unnecessary complexity when a simpler model is sufficient.
Example 2: Deep Learning — Machine Translation
Machine translation involves interpreting context, grammar, idioms, and word order across languages. Transformer-based deep-learning models use attention mechanisms to learn relationships between words in a sentence. This makes deep learning suitable for complex language tasks where manually defining all useful features would be difficult.
Machine Learning and Deep Learning Comparison
| Factor | Traditional Machine Learning | Deep Learning |
|---|---|---|
| Relationship | Uses algorithms such as linear regression, decision trees, and support vector machines | A subset of ML using neural networks with multiple layers |
| Feature definition | Often selected by people | Often learned automatically from data |
| Typical data needs | Can work with moderate datasets | Usually benefits from very large datasets |
| Computing cost | Often lower | Often higher |
| Explainability | Many models are easier to interpret | Complex models can be harder to explain |
| Example from this report | House-price prediction using linear regression | Machine translation using Transformer models |
Objective
The objective is to distinguish the major machine-learning approaches, understand the difference between machine learning and deep learning, and identify why a particular approach may be appropriate for a real-world problem.
Process
- Organized machine learning into supervised, unsupervised, semi-supervised, and reinforcement learning based on the way each approach learns.
- Reviewed the supplied comparison report and identified its two application examples: house-price prediction and machine translation.
- Compared the approaches based on data type, feature definition, data requirements, computing cost, and explainability.
- Summarized when a simpler ML model or a deep-learning model may be a better fit.
Tools and Technologies Used
This artifact was prepared as an explanatory coursework document using the supplied HTML report as its source for the ML-versus-DL comparison. The four-type overview is supplementary educational content added to complete the artifact topic. No new model was trained and no dataset or programming framework was used for this explanatory overview.
Value Proposition
Unique Value: The artifact brings together two foundational topics: how machine-learning systems learn from different kinds of training information, and how to choose between traditional ML and deep learning for a practical task. The examples make the distinctions easier to connect to real applications.
Relevance: Understanding these categories helps learners identify the nature of a problem before selecting an algorithm or model architecture. It also provides a foundation for further study of neural networks, data preparation, model evaluation, and AI applications.
Conclusion
No single approach is best for every problem. The learning approach should match the available data and the task. The supplied report emphasizes that traditional ML can be effective for structured data with known features and explainability needs, while deep learning is useful for complex inputs such as natural language when learning features automatically is valuable.
References and External Evidence
- IBM. (n.d.). Types of machine learning. https://www.ibm.com/think/topics/machine-learning-types. Supports the overview of supervised, unsupervised, semi-supervised, and reinforcement learning. IBM also discusses self-supervised learning as an additional category.
- Sarker, I. H. (2021). Machine learning: Algorithms, real-world applications and research directions. SN Computer Science, 2, 160. https://pmc.ncbi.nlm.nih.gov/articles/PMC7983091/. Provides a peer-reviewed overview of major machine-learning approaches and examples.
- Google for Developers. (n.d.). Machine Learning Crash Course. https://developers.google.com/machine-learning/crash-course. Supports the explanation of regression, model training, and neural-network fundamentals.
- Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2017). Attention is all you need. Advances in Neural Information Processing Systems, 30. https://arxiv.org/abs/1706.03762. Original research introducing the Transformer architecture, relevant to the machine-translation example.
- Wu, Y., Schuster, M., Chen, Z., Le, Q. V., Norouzi, M., et al. (2016). Google’s neural machine translation system: Bridging the gap between human and machine translation. arXiv:1609.08144. https://arxiv.org/abs/1609.08144. Research reference for neural machine translation.
Course source: Machine Learning vs. Deep Learning (2.3 Assignment), submitted course report.
Supplementary content: The four-type overview (supervised, unsupervised, semi-supervised, and reinforcement learning) was added for this portfolio artifact as an explanatory summary and was not part of the uploaded comparison report.
Artifact 3: Understanding Neural Networks Through House-Price Prediction
A simple explanation of how a neural network learns patterns from examples and estimates the price of a house it has not seen before.
Deep LearningNeural NetworksSupervised LearningRegressionIntroduction
How can a computer estimate the price of a house it has never seen? One way is to train a neural network using examples of houses whose selling prices are already known. The network learns patterns from those examples and uses what it has learned to make a prediction for a new house.
Description
A neural network is a machine-learning model made up of connected units, often described as artificial neurons. These units are organized into layers. The input layer receives information, hidden layers learn relationships in the information, and the output layer produces a prediction.
For a house-price example, inputs might include the number of bedrooms, bathrooms, and the size of the house. The network processes these inputs and produces an estimated price.
Bedrooms
Bathrooms
Square feet
Learns patterns
Estimated price
Simplified illustration. A real neural network can contain multiple hidden layers and many connected units.
Objective
Explain in simple language how a neural network learns from known examples and applies the learned patterns to make a prediction about a new example.
Process
- Provide examples: Give the model house information, such as bedrooms, bathrooms, square footage, and known selling prices.
- Process the inputs: The network passes the information through connected layers. Connections have weights that influence how strongly information contributes to the prediction.
- Make a prediction: The output layer produces an estimated house price.
- Measure the error: Compare the estimate with the actual selling price to determine how far the prediction is from the known result.
- Learn from the error: During training, the network adjusts its weights to reduce prediction error over repeated examples.
- Test with an unseen house: After training, provide details of a house not used in training to see whether the model can generalize what it learned.
Key Concepts Demonstrated
House characteristics are the information supplied to the model.
Weights influence how the network combines information when making a prediction.
The network learns by comparing predictions with known results and adjusting its weights.
A useful model should identify patterns that also help with houses it did not see during training.
Tools and Technologies Used
This artifact is a conceptual explanation of neural networks. It uses a house-price prediction example and a simplified diagram. No programming framework, dataset, or trained model was used in this explanatory activity.
Supporting Document: Open the Artifact 3 supporting document. This document provides the expanded explanation, learning objective, neural-network structure, training workflow, limitations, conclusion, and references.
Value Proposition
Unique Value: This artifact translates an abstract machine-learning topic into a familiar example. By following the path from house details to a predicted price, a reader can understand the basic roles of inputs, layers, weights, training, and prediction.
Relevance: Neural networks are a core concept in deep learning. Understanding their basic workflow provides a foundation for learning about more advanced AI/ML models and discussing how predictive systems learn from data.
Limitations and Considerations
A neural network's prediction depends on the quality and representativeness of its training data. If the examples do not reflect the houses the model later encounters, its predictions may be inaccurate. Other factors that influence house prices may also be missing from the input features. The model's output is an estimate, not a guaranteed selling price.
References and External Evidence
- Google for Developers. (n.d.). Machine Learning Crash Course: Neural networks. https://developers.google.com/machine-learning/crash-course. Explains neural-network concepts including perceptrons, hidden layers, and activation functions.
- Google for Developers. (n.d.). What is machine learning? https://developers.google.com/machine-learning/intro-to-ml/what-is-ml. Provides background on how models learn from data and make predictions, including regression.
- IBM. (n.d.). Types of machine learning. https://www.ibm.com/think/topics/machine-learning-types. Provides context for supervised learning, the learning approach used in the house-price prediction example.
Application note: These sources support the general concepts illustrated in the house-price example. The example is a conceptual explanation, not a trained or evaluated prediction model.