Skip to content
Projects
Building Encoder-Decoder Seq2Seq Transformer
(English-Spanish Translator)
I created an Encoder-Decoder Transformer from scratch using PyTorch, replicating the concepts from the influential paper "Attention is All You Need".
The model utilizes self-attention mechanisms to enhance translation accuracy, and I trained it on a large Spanish-English dataset to ensure robust performance in language translation tasks.
The results are promising.
Body Movement Detection
I developed it by first capturing video clips through a camera, each containing 15 frames per movement. I collected body coordinate data for four types of movements (right, left, jump, crouch), with 100 video clips for each movement (400 clips total). Using TensorFlow and MediaPipe, I built a CNN-LSTM deep learning model to process the data and trained it with the 400 video clips. The resulting model achieved a test accuracy of 98%, making it highly effective at classifying body movements. This model was commissioned by Mars Game Studio.
Plant Disease Detection
A web app built with the TensorFlow framework, using EfficientNet for image classification of plant leaf images. The app allows users to take a picture of a leaf and upload it to the web app, which then instantly predicts the disease type, providing an interactive and user-friendly experience through Streamlit.
Fundraising Dapp on the Ethereum Blockchain!
I developed a Fundraising DApp using JavaScript for the frontend and Solidity for the backend. Users connect their accounts via MetaMask, contribute Ether to the campaign, and check the total funds raised using the balance function (logged in the console). The Solidity smart contract ensures that only the owner can withdraw the funds, providing secure and transparent donation management. The project is split into two repositories: one for the frontend and one for the backend.
Bitcoin Price Prediction
Bitcoin Price Prediction project as a time series forecasting task, utilizing various methods, including LSTM (Long Short-Term Memory) models. By leveraging historical price data, I implemented forecasting techniques that incorporate window and horizon values to capture trends and patterns in Bitcoin’s price movements. This project explores different approaches to improve prediction accuracy, making it a valuable tool for understanding and anticipating cryptocurrency price fluctuations.
DDoS Attack Detection
DDoS Attack Detection project to evaluate the performance of machine learning-based classifiers in predicting DDoS attacks. Given the rising threat of such attacks, this project focuses on assessing classifiers like k-NN, SVM, Logistic Regression, Random Forest, AdaBoost, MLP, CNN, and Gaussian Naïve Bayes using classification metrics, training efficiency, and prediction accuracy. The evaluation is conducted from two perspectives: one using the best-selected features of the training set and the other considering all features. This approach helps identify the most effective model for detecting and mitigating DDoS attacks in real-world scenarios.
Exploring Crime Rate Trends and Forecasting Future Patterns in Toronto City
Analyzing and forecasting crime rate trends in Toronto using Police MCI (Major Crime Indicators) data and deep learning models. The project explores historical crime patterns, identifies key trends, and applies time series forecasting techniques to predict future crime rates. By leveraging deep learning algorithms, the model provides insights that can help law enforcement and policymakers make data-driven decisions for crime prevention and public safety. This research is under review for publication.
Movie Recommender System
Movie Recommender System was built u to provide personalized suggestions. By analyzing user preferences, ratings, and movie similarities, it enhances recommendation accuracy.
Traffic Signal Classification
Traffic Signal Classification model using Keras to accurately identify traffic signals from images. Using deep learning and CNNs, the model learns to classify signals based on their shapes and colors. This project enhances autonomous driving and traffic management by providing a reliable way to recognize and interpret traffic signals in real time.
MLOps Engineer
I work as an MLOps Engineer at VOsyn.ai, a voice and video company, where I manage the end-to-end machine learning deployment pipeline and MLOps using GCP and AWS. My role involves pushing fine-tuning models to GitHub, and automating workflows using GitHub Actions. The Docker images are built automatically, the pushed to GCP Artifact Registry, this is Continuous Integration (CI). For Continuous Deployment (CD), these images are deployed on GCP Cloud Run automatically, ensuring scalable and efficient model serving. Additionally, I set up Cloud Run functions for automatic triggers and serverless applications, streamlining the entire MLOps lifecycle.
Customer Service AI Agent
The Customer AI Assistant was built using LangChain, and various LLM APIs to enhance customer support and engagement. My chatbot interacts with users by answering inquiries, providing product recommendations, and assisting with troubleshooting in real time. By leveraging natural language processing and retrieval-augmented generation (RAG), it delivers accurate and context-aware responses. The assistant streamlines customer interactions, reducing response time and improving overall user experience, making support more efficient, personalized, and scalable.
TensorFlow API Reference AI Assistant
The TensorFlow API Reference Assistant was built to help users quickly find details about TensorFlow's Python API. Using Flowise, LangChain, various LLM APIs, and RAG technology, I indexed the TensorFlow documentation, allowing my chatbot to provide instant references for functions, classes, and modules. Users can simply ask about any TensorFlow symbol, and the assistant retrieves the relevant API details, making it easier to navigate the documentation. This setup streamlines the search process, ensuring users get precise and efficient responses without manually browsing through extensive documentation.
Paper Writer AI Agent
An AI-powered Research Paper Assistant was built using Flowise, LangChain, various LLM APIs, and RAG technology, to help users craft well-researched papers. My chatbot starts by asking users for their topic of interest and then suggests three trending research areas within that domain. This interactive approach helps them refine their focus before diving deeper. Once the user selects a topic, the assistant guides them through the research process, gathering relevant insights and structuring the content. In the end, it generates a complete research paper, making the process efficient, streamlined, and user-friendly.
Vacation City and Sightseeing Recommender
Web app was built using the Google PaLM LLM API and LangChain, deployed on Streamlit. Users simply select the month and continent of their travel, and the app recommends the best city to visit based on those preferences. It provides justifications for the recommendation and highlights the major sightseeing spots in the selected city, offering a personalized travel experience.
Internet Documents Research Tool
A document chatbot that lets users interact with online articles or documents. Users input URLs, click "Digest documents" for preprocessing, and then ask questions in a designated prompt section to retrieve answers from the processed content. The app uses the Google PaLM Generative AI API.
Structured Abstract Classification
An NLP model with TensorFlow was built that classifies sentences in a research paper’s abstract to make it more structured and easier to read. Using deep learning techniques, the model processes the abstract, identifying key sections such as background, methodology, results, and conclusions. By organizing the content in a clear and structured format, this project enhances readability and helps researchers quickly grasp the core ideas of a paper.
SQL Data Analysis
SQL Data Analysis project to extract and analyze website traffic and performance data from the Maven Fuzzy Factory database. The goal is to quantify the company's growth and uncover the key factors driving it. In the first project, I focused on measuring growth trends and identifying patterns in website traffic. In the second project, I analyzed marketing channels and website improvements, crafting a data-driven growth story that can be presented to executives and investors. By leveraging SQL for data extraction and analysis, this project demonstrates the impact of data-driven decision-making in business strategy.
Tweet Classification
I built a Tweet Message Prediction model for a Kaggle competition, leveraging the Universal Sentence Encoder from TensorFlow Hub, along with LSTM and 1D Convolutional models. This project analyzes tweets to understand their overall message, capturing contextual meaning and sentiment. By combining pretrained embeddings and deep learning, the model enhances accuracy in text classification and sentiment analysis.
Want to print your doc?
This is not the way.
This is not the way.
Try clicking the ⋯ next to your doc name or using a keyboard shortcut (
CtrlP
) instead.