Things I've built
Real machine learning systems designed for low latency, robust error handling, and production reliability.
Live demos
Direct inferenceBitCheck image analysis
EfficientNet-B0 · C2PA · ELAInspect an image for signs of synthetic generation and available provenance signals. Images are sent to the BitCheck model service for analysis.
Upload an image or choose a sample to run the model.
Featured projects
11 production systemsBitCheck
What it does: Detects AI-generated and manipulated media across text, images, video, and audio with multi-signal forensics and Grad-CAM explainability.
Multimodal verification API and dashboard running parallel forensics across EXIF/XMP metadata, C2PA cryptographic credentials, watermark OCR, and EfficientNet-B0 classifiers.
LockedIn AI Service
What it does: Generates structured, beginner-to-advanced learning roadmaps and curated study plans for any skill in seconds.
Autonomous learning roadmap generator leveraging Tavily Search, YouTube Data API, DeepSeek LLM, strict Pydantic v2 schemas, and SQLite caching.
AI Resume Optimizer & Career Architect
What it does: Analyzes skill gaps between candidate resumes and job descriptions with 95% parsing accuracy to recommend targeted improvements.
Automated career coach extracting competencies with ONNX INT8/FP16 quantized NER, crafting targeted cover letters with Google Gemini 2.0 Flash.
Diabetic Retinopathy Classifier
What it does: Classifies stages of Diabetic Retinopathy from retina fundus photographs with visual clinical explainability heatmaps.
Medical computer vision system using fine-tuned PyTorch EfficientNet-B0 and Grad-CAM to highlight lesions and retinal microaneurysms.
Nigerian Pidgin Next-Word Predictor
What it does: Autocompletes and predicts the next words in conversational Nigerian Pidgin English in real-time as users type.
Low-resource African NLP dual-model engine combining an LSTM recurrent neural network with statistical trigrams over a containerized FastAPI endpoint.
Brain Tumor MRI Classifier
What it does: Analyzes brain MRI scans to classify tumor types into Glioma, Meningioma, Pituitary, or healthy scans in under 200ms.
Edge-quantized medical imaging classifier diagnosing 4 brain MRI classes using transfer learning on EfficientNetB0.
CineMatch API
What it does: Delivers semantic movie recommendations based on mood, vibe, and natural language descriptions.
Semantic movie recommendation engine encoding plots with SentenceTransformers and performing sub-100ms vector similarity searches with FAISS.
Legal Document Analyzer
What it does: Reads dense multi-page contracts and legal agreements, letting users ask specific questions and instantly highlighting liable clauses.
Interactive contract analyzer extracting clauses, obligations, and risk factors from complex PDF agreements using dense semantic embeddings and cosine similarity ranking.
Flappy Bird RL
What it does: Benchmarks policy convergence, stability, and sample efficiency across four distinct RL and evolutionary control paradigms.
Reinforcement learning benchmark implementing Tabular Q-Learning, PPO, NEAT neuroevolution, and Genetic Algorithms in Gymnasium.
Bank Customer Churn Prediction
What it does: Predicts bank customer attrition risk and provides interactive parameter adjustment with feature importance analysis.
End-to-end predictive web app analyzing bank client churn risk using Gradient Boosting, handling class imbalance with interactive tuning.
Credit Card Fraud Detection
What it does: Identifies fraudulent transactions in highly skewed transaction datasets using autoencoder reconstruction error and isolation forests.
Anomaly detection system combining Isolation Forests and Deep Autoencoders with PCA and t-SNE latent space visualizers.