Aastha Singh

Product Manager

Bengaluru, Karnataka, India3 yrs 10 mos experience

Most Likely To SwitchAI Enabled

Key Highlights

Expert in multimodal AI and edge devices.
Developed scalable visual AI applications for retail.
Strong background in Natural Language Processing.

Stackforce AI infers this person is a skilled AI Engineer with expertise in multimodal AI and visual applications.

Contact

Skills

Core Skills

Image Signal ProcessingMultimodal AiArtificial Intelligence (ai)Visual AiMachine LearningComputer VisionNatural Language Processing (nlp)

Other Skills

C++Climate ModelingMathematicsVersion ControlNLP LibrariesCUDAPythonPyTorchReinforcement LearningGenerative AILarge Language Models (LLM)TableauSupply Chain OptimizationSQL

About

Welcome to my profile! Reach out to me on - aasthawork21@gmail.com In my recent experience as a Research Scholar at University of California, Berkeley, I am working on research projects in mulitimodal AI on edge devices for Natural Disasters monitoring and prediction. Here are a few things currently in my forte! * GPU acceleration * Neural networks design * High-Performance Inference Pipeline architecture * Embedded inference devices. At SparkCognition, I have been involved in developing innovative and scalable visual AI applications for retail monitoring, micro-facial emotion recognition, demographic analysis, face detection and recognition, face-ID matcher and authenticator for airports, fire detection, image enhancement and restoration, and synthetic data generation. I have leveraged my skills in Python, Machine Learning, PyTorch, NVIDIA Deepstream, to deliver high-quality and robust solutions that meet the needs and expectations of our clients. I'm always excited to chat about the latest advancements in AI and using them to better society! Hit me up if you want to talk AI tech!

Experience

3 yrs 10 mos

Total Experience

1 yr 8 mos

Average Tenure

1 yr 9 mos

Current Experience

Qualcomm

Camera Systems Modelling (Image Signal Processing) Enginner

Aug 2024 – Present · 1 yr 9 mos · Bengaluru, Karnataka, India · Hybrid

Image Signal ProcessingC++

University of california, berkeley

Research Scholar

May 2023 – Aug 2024 · 1 yr 3 mos · California, United States · On-site

Climate ModelingMathematicsMultimodal AIVersion ControlArtificial Intelligence (AI)Computer Vision+2

Sparkcognition

AI Engineer(Visual AI)

Jul 2022 – Aug 2024 · 2 yrs 1 mo · Bengaluru, Karnataka, India

Tech Stack - Edge AI, PyTorch, NVIDIA Deepstream 6.1, Docker
Projects - Retail monitoring, Micro-Facial Emotion Recognition, Demographic analysis, Face Detection and Recognition, Face - ID Matcher and authenticator for Airports, Fire Detection, Image Enhancement and Restoration, Utilising synthetic data for real world scenarios

CUDAMathematicsVersion ControlC++NLP LibrariesVisual AI+1

Integration wizards solutions

Computer Vision Engineer

Jan 2022 – May 2022 · 4 mos

Acquired by SparkCognition

MathematicsVersion ControlNLP Libraries

Nvidia

Deep Learning (NLP)

Sep 2021 – Feb 2022 · 5 mos · India

Tech Stack - Python, PyTorch, NeMo Conversational AI toolkit, SSH
Deployed on NVIDIA RIVA, Text-to-Speech and Speech-to-Text for 15 Indic Languages on multi speaker and single speaker dataset.
Trained from scratch and fine-tuned Tacotron2 with vocoders - Waveglow , HifiGAN, MelGAN
Neural Machine Translation from open sourced English to Indic Languages dataset.
Training and optimising AAYN(Attention is All You Need) transformer model with YTTM(YouTokenToMe) tokenizer.

CUDAMathematicsNLP LibrariesNatural Language Processing (NLP)

Indraprastha institute of information technology, delhi

Research Internship (AI/ML in Healthcare)

Sep 2021 – Dec 2021 · 3 mos

Halliburton

Data Science Intern

Jul 2021 – Sep 2021 · 2 mos

Tech Stacks - PySpark, Databricks, Microsoft Azure Storage Explorer, JDBC.
Multi Channel Time-Series Alignment :
Generated Lookup Table to monitor the functional activity of pumps used in Hydraulic Fracturing.
Categorised similarity between two Time Series using concepts of Dynamic Time Warping and also tested it along with Clustering algorithms.
Explored DTAN (Diffeomorphic Temporal Alignment Net) for generalising Time-Series alignment.
Time-Series Peak Detection :
Implemented functions of scipy.signals library for peak detection.
Optimised the function using Bayesian Optimization algorithm.

Ploutos

Data Science Intern

Apr 2021 – Jul 2021 · 3 mos · Remote

Tech Stacks - Python, PyTorch, Flask, REST API, AWS
Automatic Code Generation.
Developed end-to-end model by fine-tuning GPT-2 on Python libraries andevaluated the confidence score with TabNine and Kite. Deployed model using flask API.
Text summarization and document classification.
Developed end-to-end model using BERT, andspaCy(NER) after comparing performance on GPT-2, XLNet, BERT, DistilBERT, RoBERTa.
Retail Object Detection using YOLOv5 for CocaCola stock keeping units (SKUs).
The mAP (meanaverage precision) @0.5 metric was 0.75 after 150 epochs.
Built Platform documentation using Gitbook and performed AWS Bucket Pre-Processing to deploy above mentioned models on AWS Platform.