Machine Learning Course
Three course projects covering computer vision, NLP, cloud deployment, and predictive modeling. My role across the projects was project leader.
Vision
TrackMania Autopilot
NLP
Course Assistant
Cloud
Prediction Platform
ML Submission
This project focused on building a driving agent for TrackMania Nations Forever. The pipeline recorded gameplay frames, mapped keyboard actions into four classes, trained a CNN classifier, and used live screen inference to control the car.
Recorded road-focused image crops, balanced the action classes, and trained a ResNet-based classifier with fastai and PyTorch.
Ran live inference with temporal smoothing, confidence thresholds, and keyboard control to make driving behavior more stable.
NLP Challenge
The NLP project was built as a course assistant that can ingest learning materials, retrieve relevant context, and generate answers through a local LLM workflow. It combines a FastAPI backend with Redis vector search, Ollama embeddings, and LangGraph-style orchestration.
Used embeddings and vector search so answers could be grounded in uploaded course slides and documents.
Added chat storage, summaries, memory retrieval, and streaming responses to make the assistant feel more continuous.
Cloud-Serfers
Cloud-Serfers combined machine learning with deployment work. The project included data cleaning, exploratory analysis, model training, a FastAPI prediction backend, and a Streamlit frontend for testing model predictions.
Trained and tuned models with PyCaret, including XGBoost, then deployed the best housing model through an API endpoint.
Used AWS learner lab tooling, S3, SageMaker notebooks, backend endpoints, and a Streamlit interface to connect modeling with delivery.
Machine Learning
Project leader
ML, NLP, and cloud-based machine learning
Model training, NLP retrieval, cloud deployment, and team coordination