Lipie S.

Key Skills

Software

Other

Top Subject Matter

Cerebrovascular disease detection on 3D MRA imaging

Synthetic footwear image generation and style clustering

Physics-informed crack-path prediction on glass plate data

Top Data Types

3D Sensor

Document

Image

Top Task Types

Diagnosis

Fine-tuning

Classification

Data Collection

Freelancer Overview

Cerebrovascular Disease Detection using Deep Learning (University/Professor project), 2025.. Brings 3+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include PyTorch, TensorFlow, and Internal. Education includes Bachelor of Technology, Indian Institute of Technology Delhi (2021) and High School Diploma (CBSE), Delhi Public School, Greater Noida (2021). AI-training focus includes data types such as Medical, DICOM, and Computer Code and labeling workflows including Diagnosis, Fine-tuning, and Classification.

Labeling Experience

Predictive Scaling of VMSS for Azure Firewall (TPM intern), 2025.

Document

Developed an ML demo to forecast firewall demand ahead of time using a Random Forest Regressor. Drove predictive scaling specification by defining competitive landscape, business impact, and system design objectives for an Azure Firewall context. Created customer support UX prototypes in Figma to visualize predictive scaling workflows and decision flows.• Built a Random Forest-based forecasting demo to predict demand 15 minutes prior.• Defined product specification objectives including competitive landscape and business impact.• Produced Figma prototypes to communicate predictive scaling workflows.• Validated capacity adequacy improvement using reported adequacy metrics.

2025 - 2025

Physics-Informed Crack-Path Prediction (PINN + PyTorch U-Net), 2025.

Diagnosis

Developed physics-informed crack-path prediction from glass plate notch geometry and load data using PINNs. Implemented a PyTorch U-Net trained with a phase-fed PINN loss to maintain physical validity while learning fracture-related patterns. Used the framework to lower fracture energy by separating different strains through the PINN formulation.• Modeled crack paths conditioned on geometry and load inputs.• Implemented phase-fed PINN loss within a PyTorch U-Net training loop.• Ensured predictions satisfy physics constraints via PINN components.• Reported physically consistent fracture energy behavior based on strain separation.

2025 - 2025

Cerebrovascular Disease Detection using Deep Learning (University/Professor project), 2025.

Diagnosis

Conducted patch-based deep learning training on 3D MRA scans for cerebrovascular disease detection using stratified sampling. Reproduced the BRAVE-NET(3D CNN) approach and implemented a dual-patch training strategy to separately handle noise filtering and fine vessel detail learning. Evaluated model performance using Dice coefficient metrics and aligned results with published benchmarks.• Trained with patch-based sampling from 3D MRA volumes for segmentation/detection-style learning.• Implemented dual-patch filtering vs fine-detail capturing to improve vessel representation.• Used stratified sampling to balance learning across scan regions/classes.• Reported Dice coefficient of 0.86 against BRAVE-NET benchmarks.

2025 - 2025

Real-Time Suggestion Engine (Trie-based), 2024.

Classification

Built a real-time suggestion/autocomplete system for product-name search using a Trie-based approach. Integrated instant suggestion retrieval for improved user experience by storing and querying product name strings efficiently. Designed the system for fast, low-latency text suggestion behavior during user input.• Implemented Trie structure for efficient product name storage and retrieval.• Added real-time autocomplete functionality for instant suggestions.• Ensured efficient time complexity for interactive suggestion use.• Delivered an enhanced UX via relevant auto-suggestions.

2024 - 2024

Synthetic Footwear Design Data Generation (GAN), 2024.

Fine-tuning

Implemented generative modeling for synthetic footwear image generation using a GAN to create realistic footwear samples. Performed feature-based clustering and classification using PCA with K-Means and K-Nearest Neighbours to derive style clusters and shoe categories. Assessed generation quality using quality metrics and PCA visualization to validate diversity and realism.• Built a GAN pipeline for generating 30+ footwear image samples.• Applied PCA + K-Means for footwear style clustering.

2024 - 2024

Show more

Education

Work History