I'm a Computer Science student with a Math minor at Northeastern University. I'm interested in data science, data visualization, machine learning, mathematical modeling, and algorithm design.
- Studying Computer Science (Math minor) at Northeastern University
- Interests: Data Science, Machine Learning, Mathematical Modeling, Data Visualization, and Algorithm Design
- Currently learning: neural networks from first principles, deriving and verifying backpropagation by hand
- Reach me: LinkedIn and ismail.abd@northeastern.edu
- Languages: Python, Java, JavaScript, SQL
- Data/ML: Pandas, NumPy, Matplotlib, Scikit-learn, Keras
- Frameworks & Tools: FastAPI, LangChain, ChromaDB, Supabase, NetworkX, Streamlit, Git
- Visualization: Flourish, Datawrapper
- Other: Autodesk Inventor (CAD), REST APIs
| Project | Description | Tech |
|---|---|---|
| Clinical Trial Duration | Predicting how long a clinical trial will take from registration data alone. Extracted 220K trials from a 40-table medical database, then built a neural network from scratch in NumPy with hand-derived backpropagation (handwritten derivations included), verified against numerical gradients to nine decimal places. Compared eight models and deployed a demo reporting prediction ranges rather than single numbers |         |
| Soybean Lodging | Can you tell which soybean plots will fall over before harvest? Cleaned 27K plot records from 292 African variety trials where missing values were silently coded as zero across 22 columns, audited five environment tables to find four were redundant with the fifth, and built a site-history feature with a correlated SQL subquery. Logistic regression written from scratch in NumPy and verified to eight decimal places. Six models, and a negative result worth reporting: every feature is trial-level while lodging varies plot to plot | |
| Family Tree Builder | Interactive app for building a family tree and finding how any two people are related. Breadth-first search, path reconstruction, and generation layout written from scratch, with formal correctness proofs and validation that blocks impossible relationships |  |
 Huntington News Visuals |
Data visualizations for published articles at The Huntington News, Northeastern's student paper. Topics include Boston coverage, lifestyle trends, and campus coverage |   |
| AI Chat App | Full-stack chat application with persistent memory using RAG. Supports cross-conversation retrieval, file and image ingestion with OCR and captioning, and streaming responses over WebSockets |      |
- Built and numerically verified a neural network from scratch, then benchmarked it against library implementations on 220K rows of real medical data
- Found and documented data problems nothing in the source flagged, including missing values encoded as zero across 22 columns and four of five source tables being redundant with the fifth
- Mathematical modeling: hand-derived backpropagation through multiple layers and applying graph theory to genealogy
- Implemented graph algorithms from first principles, including breadth-first search with path reconstruction and formal correctness proofs
- Comfortable across the pipeline: SQL extraction from relational databases, cleaning decisions on messy data, modeling, and deployment
- Data journalism work published across multiple articles at The Huntington News
- Built a full-stack AI application working with API, database, and vector search
