Mustapha S.

Key Skills

Software

No software listed

Top Subject Matter

Medical imaging (chest X-rays) / pneumonia detection

Education analytics (student performance prediction)

Business analytics (customer churn detection)

Top Data Types

Image

Audio

Text

Top Task Types

Diagnosis

Text Generation

Classification

Action Recognition

Function Calling

Freelancer Overview

Pneumonia Detection System (CNN in PyTorch). Core strengths include PyTorch, Streamlit, and Scikit-learn. Education includes Bachelor of Science, University of Ilorin (2026). AI-training focus includes data types such as Image, Computer Code, and Programming and labeling workflows including Diagnosis, Text Generation, and Classification.

Labeling Experience

Airia — AI Email Intelligence Platform (two-agent classification + reply generation)

TextFunction Calling

Designed an AI email triage and reply workflow that classifies emails by priority and category and produces confidence-scored outputs. Built a two-agent pipeline where Agent 1 performs structured email classification and Agent 2 generates tone-matched draft replies using Gemini. Added a privacy guardrail that scans for sensitive content and blocks AI generation when protected data is detected. • Priority and category labels assigned per email for downstream actions. • Confidence scoring used to drive triage outcomes and UI labeling. • Tone-matched reply draft generation based on classification outputs. • Privacy filtering rules prevent generation for sensitive content.

2025 - Present

Speech Command Recognition System (deep CNN)

AudioAction Recognition

Trained a deep CNN model on the Google Speech Commands dataset to classify spoken commands into predefined categories such as yes, no, up, and down. Implemented real-time audio input support and built an interactive deployment for demonstrations of spoken-command recognition. Validated model accuracy on a held-out set and packaged the system for user interaction. • Audio samples were used for supervised command category classification. • CNN training for multi-class speech command recognition. • Real-time inference pipeline for live audio input. • Delivered an interactive web interface for demonstration and testing.

2024 - 2024

Customer Churn Detection System (binary classification)

Classification

Trained supervised classification models to detect customer churn and support retention decision-making via real-time predictions. Built and compared Logistic Regression and Random Forest classifiers for binary churn prediction. Deployed the solution through Streamlit to provide an accessible interface for churn inference. • Labeled examples corresponded to churn vs. non-churn outcomes. • Implemented and trained classification models (Logistic Regression, Random Forest). • Achieved high-precision performance through model evaluation. • Packaged inference in a Streamlit application.

2024 - 2024

Student Performance Prediction Web App (supervised regression)

Text Generation

Developed predictive ML workflows and deployed them in a web interface for estimating student writing scores using regression modeling. Performed feature engineering to improve accuracy and interpretability of the regression outputs. Delivered the application through Streamlit so educators could interact with model predictions. • Target variable represented student writing score estimates (regression labels). • Feature engineering and model training for supervised prediction. • Model evaluation to reach high prediction accuracy. • Deployment via a user-facing web app for prediction access.

2024 - 2024

Pneumonia Detection System (CNN in PyTorch)

ImageDiagnosis

Trained and validated a convolutional neural network (CNN) to diagnose pneumonia from chest X-ray images, focusing on reliable image inputs. Built an image verification/filtering model to confirm uploaded images were valid chest X-rays before running classification, reducing false inputs. Evaluated model performance and deployed the system for interactive use by non-technical users. • Data included chest X-ray images for pneumonia vs. non-pneumonia classification. • Implemented a preprocessing/verification step to filter invalid inputs. • Tuned and assessed CNN performance on held-out data. • Delivered an interactive app for end-users to run inference.

2024 - 2024

Education

Work History