things i've built

01

AI Health Assistant

Multimodal chest X-ray analysis system (in progress). A DenseNet121 CNN does multi-label classification for pneumonia, pleural effusion, pneumothorax, atelectasis, and cardiomegaly, since one X-ray can show several findings. Findings extracted from an uploaded PDF radiology report are compared against the model predictions, and the Streamlit interface shows per-class probabilities alongside stated uncertainty and limitations. Planned training data: NIH Chest X-rays.

Built with

PythonDenseNet121 (CNN)Multi-label classificationNIH Chest X-rays (Kaggle)PDF text extractionStreamlit
✦
02

Aztec Counselor

Course-planning chatbot that generates degree plans from an uploaded transcript. A FastAPI backend extracts the transcript data, and Amazon Bedrock converts it into structured course records and drafts the plan. Prerequisite and graduation-requirement rules are enforced with deterministic logic rather than left to the LLM. Students, requirements, completed courses, and plans are stored in PostgreSQL; transcript files go to S3.

Built with

Next.jsReactTypeScriptFastAPIPythonPostgreSQLAmazon S3Amazon Bedrock
↗
03

Traffic Detection

Runs a pretrained Ultralytics YOLO detector over 24 urban intersection images with varied lighting, weather, viewpoints, and occlusion, detecting person, bicycle, car, bus, and truck. Outputs annotated images with boxes, class labels, confidence scores, and per-class counts. Compares confidence thresholds (0.20, 0.50, 0.80) and several NMS IoU thresholds to analyze vehicle interactions with pedestrians and cyclists.

Built with

PythonUltralytics YOLOObject detectionNon-Maximum Suppression
↗
04

Line Detection

Real-time webcam pipeline in Python and OpenCV that compares three line-detection methods: Canny edge detection, probabilistic Hough transform on the Canny edge map, and least-squares line fitting on edge pixels split into left and right image regions. A four-panel display shows each method and a combined overlay for comparison under different visual conditions.

Built with

PythonOpenCVCanny edge detectionProbabilistic Hough transformLeast-squares line fitting
↗