Key Achievements:

• •Built multi-sensor data pipelines in TypeScript for real-time respiratory and postural analysis
• •Engineered algorithms for real-time respiratory and postural analysis using peak detection, FFT, and Kalman filtering
• •Built intuitive mobile app interfaces to visualize model outputs and longitudinal data

Technologies Used:

Key Achievements:

• •Built backend modules to parse unstructured medical data with OCR and AI-powered processing.
• •Engineered an AI engine with LangChain to deliver tailored health insights from longitudinal data.
• •Designed multi-agent automation for bug triaging, product feedback, and task routing.

Technologies Used:

Key Achievements:

• •Developed and deployed AI-driven workflow automation solutions (“Experts”) that translated operational pain points into scalable AI automation systems using multi-agent orchestration, persistent memory, and reusable execution pipelines
• •Designed prompt architectures, evaluation workflows and reusable agents focused on cost-efficiency and seamless end-user adoption
• •Built production-ready presets integrating enterprise collaboration tools including Slack, GitHub, and Linear to automate standups, QA coordination, project tracking and developer productivity workflows
• •Created technical to communicate platform capabilities, accelerate onboarding and support product adoption

Technologies Used:

Key Achievements:

• •Investigated hallucination detection in medical large vision-language models (LVLMs), reviewing 10+ studies to identify methodological limitations in existing hallucination benchmarks and evaluation protocols
• •Designed a benchmarking and dataset creation pipeline for evaluating hallucinations in medical AI, developing structured clinical reasoning tasks, prompt sets, and evaluation criteria to assess model factuality and reliability
• •Conducted large-scale evaluation of 11 medical LVLMs and general-purpose baselines, including GPT-4o, LLaVA-NeXT, and MiniGPT-4, across standardized medical question-answering and image-grounded reasoning tasks
• •Implemented automated evaluation pipelines for model inference, response collection, scoring, and comparative analysis, enabling reproducible experimentation across models and benchmark configurations
• •Completed the Post-Graduation Practical Training Program under the supervision of Prof. Muhammad Shafique

Technologies Used:

Key Achievements:

• •Led senior thesis research on microbiome-based machine learning for non-invasive endometriosis screening, defining the research question, objectives, experimental methodology, and evaluation framework under faculty supervision
• •Built a reproducible ML pipeline using 16S rRNA microbiome data, implementing preprocessing, feature selection, model training, cross-validation, and statistical analysis with Scikit-learn and bioinformatics tools
• •Benchmarked eight classical and deep learning models, including Random Forest, AdaBoost, and MLP, using cross-validation and a held-out test set, achieving preliminary performance of F1: 0.87 and ROC-AUC: 0.95
• •Applied feature selection methods to reduce noise and multicollinearity, increasing model accuracy by 5–20%
• •Identified potential genus-level microbial biomarkers associated with endometriosis through differential abundance analysis
• •Manuscript presented and published at the IEEE International Conference on Healthcare Informatics (USA) and ACM Conference on Bioinformatics, Computational Biology, and Health Informatics (Italy)

Technologies Used:

Key Achievements:

• •Reduced portal response time by 4% by diagnosing and resolving system bottlenecks.
• •Streamlined quarterly report reviews, cutting time by 7% with automated SQL and Python workflows.
• •Ensured 100% compliance across 5+ national systems by auditing and validating datasets.

Technologies Used: