Dev Patel

Machine Learning Engineer Aspirant

[email protected]

Hello, I am

Dev Patel.

2

Machine Learning
Projects

1

Deep Learning
Project

2

GenAI
Projects

About Me

I'm Dev Patel, an aspiring Machine Learning Engineer with a strong foundation in Python, SQL, and Machine Learning fundamentals. I'm passionate about building AI/ML systems and leveraging data to solve real-world problems. I post here the projects other than in my resume

🛠 Tools and Technologies:
🔹 I use Python to build and train machine learning models.
🔹 I’m skilled in SQL for organizing, managing, and querying structured data.
🔹 I have hands-on experience Jupyter Notebooks, Pandas, Matplotlib, and Scikit-learn for data analysis and model building.
🔹 I'm familiar with tools like MLflow for tracking experiments and managing ML pipelines.

📌 Areas of Expertise:
✅ Developing models that solve real-world problems and provide data-driven insights.
✅ Automating machine learning workflows to improve efficiency and performance.
✅ Creating clean and easy-to-interpret visualizations to support decision-making.
✅ Applying best practices in reproducibility and model lifecycle management.

Key Skills

Python

Sql

Machine Learning

My Projects

Health Insurance Prediction

Domain/Function: HealthCare

Credit Risk Classification

Domain/Function: Finanace

ResNet18 Implementation

Domain/Function: CNN (Image Classification)

RAG Using LangChain

Domain/Function: RAG

E Commerce Chatbot

Domain/Function: E Commerce

Awards & Certificate

Python: Beginner to Advanced For Data Professionals

SQL for Data Science

Math and Statistics For AI, Data Science

Master Machine Learning for Data Science & AI: Beginner to Advanced

Deep Learning: Beginner to Advanced

Natural Language Processing

Gen AI to Agentic AI with Business Projects

Let's Connect

Feel free to get in touch with me. I am always open to discussing new projects, creative ideas or opportunities to be part of your visions.

Mail Me

[email protected]

https://www.linkedin.com/in/dev-patel-data-nerd/

GitHub

https://github.com/devpatel45

Machine Learning Project

Health Insurance Prediction

Domain/Function: HealthCare

This project predicts the health care premium on the inputs provided through a streamlit UI application then is processed and the result is displayed.

View Project
on GitHUB

View Project
on

This is the Streamlit UI I build for the machine learning project. Through this UI the input data is passed to the model and the processing code.

This is a screen shot of the code which uses the model and gives and output after that the output is passed and showed on the streamlit UI.

Project Details

Health Premium Prediction (Regression)
Project Overview:
The Health Premium Predictor is a regression-based machine learning model that estimates an individual’s health insurance premium based on input features such as age, gender, BMI, smoking status, and the number of children. The goal is to predict premiums based on these factors, helping users understand potential costs.

It gave me an understanding of how a model is deployed and used for predictions.

Machine Learning Project

Credit Risk Classification

Domain/Function: Finanace

This ML Project predicts the defaulting possibility of a debtor, predicts his credit score and gives him a rating. The app UI is made using Streamlit.

View Project
on GitHUB

View Project
on

This screenshot showcases the user interface of a credit risk prediction app developed using Streamlit.

This screenshot shows a Python script used for loading a saved machine learning model and preparing input data for prediction.

Project Details

Developed an end-to-end machine learning application to evaluate credit risk by predicting the probability of loan default, estimating a credit score, and assigning a risk rating (e.g., Excellent, Good, Fair, Poor).

This Project provided me with experience of how ML classification model are used for generating the output and how things work
The Project involved lots of feature engineering and data cleaning, exploration and so much of experimentation which gave me an idea of how the work is being done in the industry.

Overall, this project provided me with lots of understand of the small but important thing like data scaling and preprocessing' s impact on the model.

Deep Learning Project

ResNet18 Implementation

Domain/Function: CNN (Image Classification)

I did this project to get the understanding of the workings of the ResNet18's Architecture and got a Brief understanding of the Architecture of the ResNet18

View Project
Engagement

View Project
on GitHUB

The Image Class Predicted by the model

Grad-CAM overlay

Project Details

I’m excited to share my latest deep learning project where I implemented the classic ResNet18 architecture from scratch — a powerful convolutional neural network known for its residual blocks that enable training very deep networks effectively.

But I didn’t stop there! To make the model’s decisions transparent and interpretable, I integrated Grad-CAM (Gradient-weighted Class Activation Mapping), a cutting-edge technique that produces heatmaps highlighting the exact regions in an image that influence the model’s prediction.

What I did:
Built ResNet18 with residual connections for efficient training of deep networks

Trained the model on a custom car damage dataset it was very small dataset (2300)

Implemented Grad-CAM to visualize model’s attention on input images

Developed an interactive Streamlit web app for users to upload images, get predictions, and see Grad-CAM heatmaps in real-time

Ensured modular, clean, and production-ready code suitable for deployment

Project video

GenAI Project

RAG Using LangChain

Domain/Function: RAG

This project we can pass the URLs which will be processed by Langchain and then stored in a chromadb vector database through which the Llama LLM can get the info

View Project
Engagement

View Project
on GitHUB

This Screen Shot shows the info from the articles

This Screen shot shows if the Access is denied what will the LLM give an output

Project Details

📚 Langchain + LLaMA + ChromaDB Project (with Streamlit UI)
This project allows users to input URLs containing useful content—such as articles, blogs, or documentation—and then ask questions about the content using an LLaMA.

🔧 How It Works
User Inputs a URL
Using the Streamlit UI then Content Loading & Processing (Langchain)
The URLs are processed using Langchain, which fetches the webpage content and cleans it. The content is then chunked into smaller pieces for better handling by the LLM.

Embedding & Storage (ChromaDB)
Each content chunk is converted into vector embeddings using a pre-trained embedding model (like OpenAI, HuggingFace, etc.).
These embeddings are stored in ChromaDB, a vector database that supports fast similarity search.

Question Answering with LLaMA
When a user asks a question, Langchain performs a vector search in ChromaDB to retrieve relevant info

Project video

GenAI Project

E Commerce Chatbot

Domain/Function: E Commerce

I have used a small FAQ data and shoes product data for showing the chatbot. This chatbot uses LLM and can Intelligently identify that the question is product related or FAQ related and answers accordingly

View Project
Engagement

View Project
on GitHUB

Project Details

For this chatbot demo, I used a small dataset containing product information (focused on shoes) and a set of frequently asked questions (FAQs). The chatbot is powered by a Large Language Model (LLM) and is designed to intelligently classify the user's intent based on their query.

It uses Semantic Routing to determine whether a question is related to the product catalog (e.g., "Do you have red shoes under $100?") or to a general FAQ (e.g., "Can I return my order after 10 days?"). Depending on the identified intent:

If it's product-related, the chatbot queries the SQLite3 database to retrieve relevant product details.

If it's FAQ-related, it performs a semantic search through ChromaDB, a vector store containing embedded FAQ content, and uses the retrieved context to answer.

Finally, the appropriate context and the user query are passed to the Groq-hosted LLM (like LLaMA or Mistral) to generate a natural, conversational response.