Hysam
Amran

Client-ready remote-sensing AI: weakly supervised U-Net/ResNet34 that segments buildings from aerial imagery using sparse click labels — cuts annotation cost while keeping competitive mIoU.

Production-style multimodal system fusing eye tracking, facial emotion, and keystroke dynamics for real-time cognitive load scoring — FastAPI + Next.js dashboard with <200ms streaming.

Domain RAG assistant over clinical anaesthesiology texts — local embeddings, FAISS retrieval, and Gemini answers with a FastAPI backend clients can white-label for healthcare knowledge bases.

End-to-end retrieval-augmented generation stack for private document Q&A — ideal for companies that need grounded answers from their own manuals, policies, or research PDFs.

Benchmark suite for business forecasting: ARIMA, SARIMA, LSTM, Prophet, ANN, and hybrids — plus a Node dashboard to ship charts stakeholders actually use.

Star-schema warehouse with mesh-join ETL linking products, customers, and transactions — the backbone pattern for retail BI, OLAP reporting, and near real-time ops metrics.

Dockerized ML service with Jenkins CI and prediction API — shows how I take models from notebook to deployable endpoints with repeatable builds.

Retail scraping + insight notebooks that turn Amazon listings into structured datasets for pricing, assortment, and competitive intelligence work.

Selenium + BeautifulSoup pipeline for JavaScript-heavy sites — exports clean JSON/CSV so clients get reliable data even when static scrapers fail.

Full-stack academic platform (Admin / Teacher / Student) with JWT auth, grading, CLO/PLO outcomes, and PostgreSQL — proof of complex role-based product delivery.