MOHAMMAD
SALEM

Building AI pipelines to process pharmaceutical vendor files using OCR (PyMuPDF, Tesseract, PaddleOCR). Currently implementing text extraction via bounding boxes and date normalization logic.

Developing a RAG-based document retrieval system with LlamaIndex, optimized with chunk tuning and open-source LLMs (Mistral, Phi-2).

Conducting end-to-end evaluation of the document intelligence system—benchmarking OCR accuracy, RAG retrieval quality, and routing performance.

Developing a machine learning classifier to identify pre-seizure physiological patterns by analyzing large-scale EKG/ECG datasets from wearable monitors.

Engineering and training neural networks on multi-dimensional time-series data to detect subtle autonomic shifts that precede clinical seizure onset.

Optimizing model performance for real-time inference on smartwatch hardware and edge devices.

Collaborated with the Chief of Obstetrics to translate clinical protocols into app features like BP triggers, observation timers, and treatment windows.

Co-developed core app logic for role-based notifications, enabling real-time coordination between nurses and residents.

Delivered a winning pitch and live demo to hospital stakeholders, focusing on reducing 'door-to-needle' time.

Launched on iOS App Store, handling all privacy compliance, monetization, and analytics infrastructure.

Architected a local-first content pipeline using Whisper for transcription and RAG for Q&A, allowing users to 'chat' with their audio.

Built a full-stack solution with React Native, TypeScript, and Supabase, implementing complex features like vector search and edge functions.

Directed the entire product lifecycle from user research to UI/UX design and technical specification.

Designed a multi-provider waterfall enrichment process, achieving >96% verified emails and improving data quality significantly.

Reduced outreach and list-building costs by >45% by replacing manual sourcing with automated data-driven systems.

Standardized targeting processes by creating reusable filtering templates and documentation for future campaigns.

Cleaned and standardized client databases to unlock new marketing channels and improve customer segmentation.

Generated geographical heat maps of sales performance, identifying 7+ high-value market expansion opportunities.

Developed an XGBoost machine learning model to predict future purchase categories, deploying it via a Streamlit dashboard.

Managed end-to-end sign production projects while simultaneously leading data visualization efforts.