This is a  project in Autonomic Software and Systems class in my first year of my master's degree. I used CARLA 0.9.14 to complete milestone 1,2 & 3. 

Milestone 1 - setting up carla and implementing executor

Milestone 2 - implement route calculation

Milestone 3 - implement obstacle avoidance

 I was part of the ML developer team to provide  ML models that will help Tano Secure Inc. improve Tiaga system.

Image classification

First model: Using an image classification model. It classify whether the image is healthy leaf or a diseased one. This model can be used when creating a mobile app.

Image Segmentation:

Second model: Semantic image segmentation is to label each pixel of an image with a corresponding class of what is being represented. This model can be used  using CCTV camera.

Exploring the intersection of machine learning, trading, and data-driven decision-making.

Dataset:
The dataset was sourced directly from NinjaTrader, containing minute-level OHLCV (Open, High, Low, Close, Volume) data. This granular market data provided the foundation for building a model capable of recognizing subtle short-term price movements.

Preprocessing:

• Data was cleaned, time-aligned, and normalized using Min–Max scaling to stabilize the training process. Missing values were handled with forward-filling, and sequences were constructed into 240-minute windows for time-series learning. 

Model Training:

• I implemented a custom Transformer architecture in PyTorch optimized for financial time-series forecasting. Training incorporated early stopping to prevent overfitting, along with validation monitoring to ensure robust performance.

• After 42 epochs, the model achieved a validation loss of 4.57 and a test MAE of 40.9, demonstrating a consistent ability to capture short-term trends. 

Prediction:

• The model’s forecasts were plotted alongside real MNQ prices to visualize prediction accuracy and momentum alignment. This experiment lays the groundwork for developing a TradingView indicator in the future, potentially connecting the Transformer’s predictions to real-time trading signals through a custom API.

• This project reflects how I enjoy transforming complex data models into actionable trading tools, combining my interests in technology, finance, and creative problem-solving.

Google Colab  CODE

Dataset:
The Kaggle dataset contains thousands of labeled email messages, marked as “spam” or “ham” (not spam).

Preprocessing:

• Cleaned the text by lowercasing and removing unnecessary characters.

• Converted the email text into numerical features using vectorization (Bag-of-Words / TF-IDF).

Model Training:

• Trained a machine learning classifier in Python using scikit-learn.

• The model learns to distinguish spam from non-spam messages based on word patterns.

Prediction:

• When given an email message as input, the model outputs:

  • 1 → Spam

  • 0 → Not Spam

Google Colab  CODE