Clinical AI
Portfolio

Clinical AI & Explainable Systems

Evidence-grounded RAG architectures and explainable vision pipelines built with rigorous citations, transparency, and clinical utility.

Clinical RAG & Decision Support KDIGO 2026 · PubMed

Nephrology Clinical Evidence Assistant

Dual-source clinical decision support assistant that queries clinical practice guidelines (KDIGO 2026 AKI/AKD) and peer-reviewed literature with table preservation and true printed page grounding.

  • Dual Evidence Sources: Clinicians can search guidelines, publications, or both; the system explicitly distinguishes guideline recommendations from research trial findings.
  • Preserved Staging Tables: Extracts complex tables via pdfplumber as intact Markdown ([TABLE START]...[TABLE END]) to keep AKI/AKD staging grids aligned.
  • True Printed Page Numbers: Uses PyMuPDF (fitz) to extract logical page numbers matching physical printed guideline pages.
  • Retrieval & Reranking: Multi-query synonym and German medical term expansion (Qwen-30B), diverse MMR search over dual FAISS vector stores, and local FlashRank cross-encoder reranking.
LangChain FAISS FlashRank Qwen-30B e5-mistral-7b pdfplumber PyMuPDF Streamlit
Active RAG Dev & Eval Discuss Architecture →
Computer Vision & XAI FastAPI & Docker

Explainable Deepfake Forensic Detection

Multi-model fusion classifier paired with Grad-CAM visual heatmaps, enabling transparent visual forensic verification in high-stakes environments.

  • Explainability Focus: Solves black-box opacity by generating visual saliency heatmaps showing exactly which spatial regions influenced the classification.
  • Model Ensemble: Multi-architecture fusion ensemble combining ResNet, EfficientNet, and DenseNet, reaching 99.97% accuracy on Celeb-DF v2.
  • Interactive Deployment: Containerized FastAPI backend with an interactive Streamlit UI for real-time visual inspection and forensic auditing.
PyTorch Grad-CAM (XAI) FastAPI Docker Streamlit
Containerized Deployment View Tech Stack →
Upcoming Initiative Bio-Signal Modeling

DeepSleep Initiative

Upcoming technical initiative focused on advanced sleep data analysis and predictive bio-signal modeling.

  • Scope: Engineering computational approaches for nocturnal physiological and biometric signal analysis.
  • Status: Research planning and data exploration phase; technical architecture and outcomes to be shared upon completion.
Time-Series Bio-Signals PyTorch
Upcoming Research In Planning

Interested in discussing these pipelines?

I am actively seeking Working Student and Junior ML Engineering roles in Clinical AI and MedTech.

Start a Conversation →