Ananya Singhal - Experience

Aug 2024 Present

GE Healthcare

AI Researcher Bangalore, India

Enhanced analytical denoising methods for high b-value diffusion data by implementing a gradient-based kernel allocation system that automatically preserves anatomical structures, applying targeted smoothing based on regional complexity for superior detail retention.

Python Diffusion MRI (dMRI) DIPY TensorFlow

May 2024 Aug 2024

Research Assistant Delhi NCR, India

Developed GraphNeuralMRI framework for early Alzheimer's disease detection using diffusion MRI data, achieving 98.9% accuracy in CN vs MCI classification and 97.99% in CN vs AD classification.
Engineered histogram-based connectivity features from brain networks derived from tractography fiber counts, significantly outperforming traditional machine learning approaches (68-79% accuracy).
Implemented and compared multiple graph neural network architectures (GCN, GAT, ARMAConv) in a transductive learning setup to capture subtle connectivity alterations characteristic of neurodegeneration stages.

Python GNNs Dipy

Jan 2024 Aug 2024

SWE Intern Bangalore, India

Developed a ML model to estimate order fulfillment status and predict the likelihood of order sales based on configurations.
Utilized SQL Server to analyze data and reduce idle inventory by identifying configurations less likely to sell.

SQL Server Pandas Scikit-learn Python

May 2023 Aug 2023

Summer Intern Bangalore, India

Architected and deployed an internal knowledge management system using GPT4ALL, LangChain, FAISS vector database, and PyPDF2 that enables employees to query 100+ company documents through natural language, reducing information retrieval time.
Implemented an end-to-end document processing pipeline with sentence-transformers for embedding generation, Chroma DB for efficient vector storage, and a Flask/FastAPI backend.

LangChain GPT4ALL/LLMs PyTorch PyPDF2/PDF Processing Python

October 2022 Aug 2023

Undergraduate Research Fellow Bangalore, India

Investigated the potential of Transformer based deep learning techniques to accelerate the processing of diffusion tensor imaging (DTI) measures and improve the early diagnosis of Alzheimer's disease (AD) using sparse data on ADNI dataset
The proposed model leverages the attention mechanism to generate quantitative measures of fractional anisotropy (FA), axial diffusivity (AxD), and mean diffusivity (MD) using 5 and 21 diffusion directions, making it useful for clinical diagnosis through reduced scanning time of more than half.

PyTorch NiBable SimpleITK Tractography FSL (FMRIB Software Library) MRtrix3 Dipy (Diffusion Imaging in Python) ANTs (Advanced Normalization Tools)

July 2022 October 2022

Research Intern Bangalore, India

Contributed to a multimodal biometric authentication system using finger vein and fingerprint data, enhancing spoof resistance through feature-level fusion. Supported preprocessing (CLAHE, ROI extraction) and deep learning-based verification using Siamese networks, achieving 95%+ accuracy.

TensorFlow OpenCV scikit-image Siamese Networks CLAHE (Contrast Limited Adaptive Histogram Equalization) CNN/ResNet NumPy/SciPy

May 2022 Aug 2022

Academic Intern Singapore

Designed and developed a system for real-time violence detection through surveillance cameras by optimizing LRCN model that uses both CNN and LSTM. The model is trained on a novel dataset that consists of 150 videos each for 'Fight' and 'noFight'.
Created a web-app using Steamlit that sends notifications and generates the map depicting exactly where the Fight is detected based on the database of the location of the surveillance cameras.

TensorFlow OpenCV LRCN (CNN + LSTM)/ Streamlit Google Maps API

May 2021 December 2021

Research Intern @CVIT Lab Hyderabad

Worked on a task of mitigating biasing of facial Recognition in Computer Vision under Dr. Anoop Namboodiri.
Created a novel publicly available dataset of 30,000 face images pertaining to famous Indian personalities (identified by similarity score), along with a web annotation system.
Mapped different races in terms of the average difference between two observable quantities, computed the error rate of facial recognition system in different parts of the space, and found a correlation between the accuracy and the location of the space.

PyTorch OpenCV Dlib MTCNN