genomic sequence
SYSTEM_INIT: READY

Advancing Health
through Applied AI

PhD Candidate specializing in practical AI applications for clinical healthcare. Developing robust systems for real-world medical impact.

// CORE_COMPETENCIES

Research Interests

Current focus: Translational AI
for Clinical Systems
clinical_notes

Clinical AI Integration

Developing models that seamlessly bridge the gap between bench research and bedside clinical workflows. Focus on human-in-the-loop systems that enhance practitioner decision-making without disrupting established medical protocols.

DATA_STREAM_01
monitoring

Predictive Analytics

Leveraging deep learning for real-time patient risk stratification and preventive care optimization in acute care settings.

visibility

Medical Imaging

Designing robust computer vision systems for automated, high-precision diagnostic support across MRI, CT, and histopathology.

Real-World Practicality

Investigation of model robustness, interpretability, and ethical deployment in diverse hospital environments with heterogeneous data sources.

PyTorch DICOM FHIR TensorRT

Featured Projects

ecg_heart
LVEF: 55% — NORMAL
BSE_DIASTOLIC: GRADE_I
HF_SUBTYPE: HFpEF
AV_VMAX: 1.2 m/s
[ PROJECT_01 ] 2024 - PRESENT

ECHO PARSER / ECHO REPORT CLASSIFIER

A Python tool that parses echocardiogram PDF reports and applies automated classification logic based on ASE and BSE guidelines. Extracts LVEF, diastolic function, heart failure subtypes (HFrEF/HFmrEF/HFpEF), valve disease severity, and LV geometry — outputting an annotated Excel dataset ready for ML training.

Python pdfplumber ASE/BSE Guidelines Tkinter GUI
nutrition
MODEL: Qwen2-1.5B-Instruct (4-bit)
DOMAIN: Malaysia MoH Maternal Guidelines
BACKEND: FastAPI + LangChain
VRAM: 4GB (RTX 3050)
[ PROJECT_02 ] 2024 - PRESENT

CLARA / MEDICAL NUTRITION AI

A domain-specific medical nutrition AI built by fine-tuning Qwen2-1.5B-Instruct on Malaysia Ministry of Health maternal dietary guidelines using Unsloth with 4-bit QLoRA. Deployed as a FastAPI REST API with LangChain integration, designed to run on consumer-grade hardware (RTX 3050, 4GB VRAM).

Qwen2-1.5B Unsloth QLoRA FastAPI LangChain
menu_book
iOS (SwiftUI) + Web (Next.js 16)
Browser Extension (MV3)
DOI / ISBN / arXiv Scanning
Supabase + Zotero Export
[ PROJECT_03 ] 2024 - PRESENT

SCHOLARSYNC / PAPER QUEUE MANAGER

A cross-platform scholarly reading queue manager with an iOS app (SwiftUI), Next.js web dashboard, and Manifest V3 browser extension — all sharing a Supabase backend. Scans DOI, ISBN, and arXiv IDs, auto-tags papers with capture location, and exports to Zotero in BibTeX/RIS/CSV formats.

SwiftUI Next.js Supabase Vision OCR Zotero API
edit_note
ADIME FRAMEWORK
Assessment → Diagnosis
Intervention → Monitoring
Fine-tuned Scribe Model
[ PROJECT_04 ] 2024 - PRESENT

SCRIBESTREAM / ADIME DIETITIAN AI SCRIBE

A fine-tuned AI scribe built for clinical dietitians, structured around the ADIME documentation framework (Assessment, Diagnosis, Intervention, Monitoring & Evaluation). Automates clinical nutrition charting to reduce documentation burden in hospital settings.

ADIME Framework Fine-tuning Unsloth Clinical NLP

// SYSTEM_INFRASTRUCTURE

Homelab Setup

[HOME_SERVER_01]
COMPUTE
CPURyzen 5 5600H
GPURTX 3050
VRAM4GB
STORAGE
INTERNAL512GB SSD
EXTERNAL1TB SSD
OSUbuntu Server
Tunnel Topology
WAN
CLOUDFLARED
UBUNTU_SERVER
SERVICES
CLOUDFLARED // UBUNTU_SERVER

A personal research server running Ubuntu Server, used for AI model training, local inference, dataset storage, and self-hosting — all tunnelled securely to the internet via Cloudflare without port forwarding.

  • RTX 3050 (4GB VRAM) / QLoRA fine-tuning & local inference
  • 512GB + 1TB SSD / Dataset & model storage
  • Cloudflare Tunnel / Zero-trust self-hosting, no exposed ports

// OPEN_SOURCE_MODELS

CONTRIBUTING_TO_COMMUNITY
Hugging Face

LLM Fine-tuning

Adapting Llama-3 and Mistral architectures for clinical reasoning using LoRA and QLoRA techniques on local hardware.

PEFT Deepspeed
biotech

Medical Imaging

Training vision transformers (ViT) and CNNs for specific diagnostic tasks using open-source PyTorch frameworks and TorchVision.

MONAI PyTorch
developer_board

Local Inference

Optimizing model deployments using vLLM and TensorRT-LLM for low-latency inference on consumer-grade hardware.

vLLM GGUF
Check out my contributions

I regularly release quantized weights and training scripts for medical-domain adaptations of popular open models.

Visit HuggingFace Profile

Let's Collaborate

Interested in healthcare AI research, practical clinical deployments, or speaking engagements? My terminal is always open.