Open to opportunities

Iram Kamdar

@iramkamdar

Message

Machine learning engineer specializing in multimodal AI, retrieval systems, and clinical ML pipelines.

United States

Message

What I'm looking for

I seek roles building production-ready multimodal ML systems and retrieval/LLM pipelines with low-latency constraints, collaborative teams, and opportunities to impact clinical or product outcomes.

I am a machine learning engineer with hands-on experience building retrieval systems, fine-tuned LLM deployments, and multimodal pipelines that reduce latency and infrastructure costs.

At WebAI I migrated JSON retrieval to Qdrant with HNSW indexing and scalar quantization, cutting retrieval latency from 39s to 2.9s and improving hit rate, and I built LoRA fine-tuning scripts to enable cost-efficient domain-specific LLMs.

As a graduate researcher I engineered predictive modeling pipelines for multimodal imaging, extracted 100+ radiomic features, and improved ranking performance via regularized Cox models; I also implemented clinical NLP using GPT-4 to structure EHR notes for downstream analysis.

My background includes applied NLP, computer vision, OCR pipelines, and production tooling (PyTorch, Hugging Face, Docker, Azure), and I focus on reliable, low-latency deployments and reproducible ML systems that directly support clinical and product goals.

Experience

Work history, roles, and key accomplishments

Current

Machine Learning Engineer

Current

WebAI

May 2025 - Present (5 months)

Reduced retrieval latency from 39s to 2.9s and improved hit rate 0.93→0.96 by migrating retrieval to Qdrant with HNSW, scalar quantization, and multi-vector pooling; built LoRA fine-tuning for LLaMA 3.1 8B to enable cost-efficient domain-specific LLM deployments and implemented low-latency multimodal image verification pipelines.

Qdrant HNSW Vector Search LLaMA PyTorch Inference Engineering

Current

Graduate Researcher

Current

Computational Biomarker Imaging Group

Sep 2024 - Present (1 year 1 month)

Engineered a predictive pipeline segmenting multimodal scans into 3D supervoxels and extracting 100+ radiomic features, and improved ranking metric from 0.67 to 0.75 via regularized Cox models benchmarked against Random Survival Forests; implemented GPT-4 clinical NLP to extract disease-activity signals from EHR notes.

Medical Imaging Radiomics SLIC Supervoxels Cox Regression Random Survival Forests GPT 4 Python Statistical Modelling

Data Scientist

Gnowit Inc

Jun 2023 - Mar 2024 (9 months)

Leveraged Llama-2 to generate article summaries, reducing curation time from hours to minutes; implemented named entity extraction and deduplication using SpaCy/NLTK and clustered events with DBSCAN/HDBSCAN/Spectral Cl, and parallelized OCR preprocessing with OpenCV and Tesseract across 5,000+ documents.

LLaMA 2 NLP SpaCy NLTK Clustering (DBSCAN HDBSCAN)OpenCV Tesseract Python data preprocessing

MITACS Globalink Researcher

Université du Québec en Outaouais

Jun 2023 - Aug 2023 (2 months)

Developed a brain tumor detection pipeline using YOLOv5 and dynamic thresholding, achieving >98.5% MRI classification accuracy and reduced analysis time by 20% via advanced preprocessing (Otsu thresholding) and personalized sensitivity tuning.

Otsu Thresholding Model Evaluation Python Computer Vision MRI