About
My Introduction
With a relentless curiosity for solving real-world challenges through technology, I specialize in pushing the boundaries of machine learning and artificial intelligence. As a Master's student at the University of California, San Diego, I focus on creating impactful solutions in areas like computer vision, natural language processing, and generative AI. My professional journey spans roles at industry leaders such as Wells Fargo, Samsung R&D, and IIT Bombay, where I honed my skills in developing scalable AI systems, conducting groundbreaking research, and delivering measurable results. I am passionate about bridging the gap between theoretical innovation and practical application, contributing to advancements that shape the future of AI.
Skills
My technical & miscellaneous skillsAI & Data Science
3+ Years XPComputer Vision
Natural Language Processing
Generative AI
Probability & Statistics
Programming
3+ Years XPPython
C & C++
Computing
1+ Years XPAWS
Distributed Computing
Misc
2+ Years XPGit
Linux
Experience
My journey in Academics & Professional Internships
Masters of Science in Computer Science
University of California, San Diego GPA : 3.92/4.0Spring 2025 :
CSE 237D : Embedded Systems Design Project
CSE 253 : Machine Learning for Music
COGS 209 : Statistical Machine Learning for Social Sciences
Winter 2025 :
CSE 251A : ML: Learning Algorithms
CSE 251C : ML: Machine Learning Theory
CSE 291J : Fair and transparent ML
Fall 2024 :
CSE 202 : Algorithm Design and Analysis
CSE 256 : Statistical Natural Language Processing
CSE 258 : Recommender Systems and Web Mining
B.Tech - Computer Science and Engineering
SRM Institute of Science and Technology , KTR Campus CGPA : 9.56 (Final)As an undergraduate student researcher :
- Co-authored two impactful research papers on Gait Speed-Based Individual Recognition using thermal images and presented findings at IEEE Explore International Conferences.
- Published a study comparing DenseNet201 and MobileNetV2 for medical image analysis using a 2-D Convolutional Neural Network.
- Collaborated with Dr. Vijaya K and Mrs. Nithyakani P at SRM IST, bridging the gap between computer science and healthcare.
Class XII
Maths, Physics, Chemistry, Computer Science | ISC , IndiaClass X
ICSE , India
Graduate Student Researcher
UC San Diego Health • La Jolla, CA - Refactored 5k+ LOC MATLAB neuroimaging code into modular Python pipelines, reducing runtime by 20%.- Applied scalable feature selection and nested cross-validation to 10k+ MRI scans, ensuring reproducibility.
- Developed predictive models (Logistic Regression, SVM) on structural/diffusion MRI using multi-modal data.
- Collaborated with neuroscientists to integrate mixed-effects models into production-ready research pipelines.
- Researcher with the BRIDGE Lab; collaboration with DAIC (ABCD Study) on multimodal imaging.
Research Intern (IRCC)
Indian Institute of Technology Bombay - Incorporated Computer Vision techniques for affective engagement to analyze peak emotions in 100+ participant interactions, prepared descriptive statistics on a study log dataset of 80,000+ entries, and applied open-source valence-arousal models.- Annotated over 30,000 video frames and utilized YOLOv8 for detecting cognitive engagement in a classroom environment using the ICAP framework, leveraging DGX machines for video analytics.
Intern Analyst (CEDA)
Wells Fargo - Developed a Python-based data extraction tool using regex, text parsing, and NLP, reducing manual Handbook processing time by 70% and improving structured data accuracy by 30%.- Applied Latent Dirichlet Allocation (LDA) topic modeling on FHA Mortgage policies, achieving a 20% increase in classification accuracy and a coherence score of 0.78.
Research Intern (PRISM)
Samsung R&D Institute India-Bangalore - Conducted forecasting on a French bakery dataset, achieving a 15% reduction in forecasting mean absolute errors using algorithms like SARIMA, Holt-Winters, LSTM with training windows from 30-180 days.- Executed time series analysis exploring mean and standard deviation trends, seasonal effects, and confirmed data stationarity with the Augmented Dickey-Fuller (ADF) test, with a p-value < 0.05 for 95% confidence.
Academic Intern
National University of Singapore - Leveraged Spacy for Named Entity Recognition and Deep neural network to develop a robust Resume Parser and multi-label classifier, achieving an impressive precision rate of 96% for the classification of developer resumes.- Pioneered integration of cutting-edge Deep Learning optimization techniques; achieved a 10% increase in classification precision and streamlined workflows, elevating team productivity by 15%.
Projects
My works, projects & contributions
Benchmarking Music Conversational Recommendation
Built a novel multi-modal dataset by combining Reddit dialogues with YouTube audio (FFmpeg, Python). Implemented entity linking and benchmarked LLMs (Qwen-2.5-7B) using PyTorch Transformers.
Fire-Ready Forests Data Challenge (Winner)
Processed ecological and spatial data with missing-value handling and feature engineering for species classification. Built ensemble models achieving 96.9% test precision on the National Data Platform; results presented at ICPP 2025.
View Repo
BlogPod
Pipeline using LLaMA 3.1 (8B/70B/405B) to convert podcast transcripts into structured blogs via prompt engineering. Evaluation with Gemini-1.5-Flash on clarity, tone, and engagement.
View Repo
TinyML Owl Vocalization
Built sub-0.5MB CNNs in PyTorch for owl call classification; quantized and exported to ONNX/TFLite. Deployed on STM32 with X-Cube.AI and TFLite Micro for real-time, low-power acoustic monitoring (with E4E & San Diego Zoo Wildlife Alliance).
View Repo
Asset Bubble
Project helps the investors using a predictive analytics model or providing actionable analytical insights to this problem. The output is the confidence score on whether the stock market will crash in 2022 or not and if it does find an approximate time interval in which this can possibly occur.
View
Alveoli
Chest X-Ray reports in a matter of seconds.We have created a secure Chest X-Ray image classification based website and app that can detect Covid-19, Pneumonia and Tuberculosis. Deep Learning has been used to detect the disease by using a Convolutional Neural Network(MobileNetV2) which performs classification
View Repo
ResPas
This project is aimed at building a multi-label classification model using the principles of Deep Learning and NLP to effectively classify users into job categories given their resume.
View
BTC Prediction
In this Deep Learning Application, we have used the price data for Bitcoin to forecast its price in a specified future window. We have used the Tensorflow and Keras APIs to build a stacked LSTM model.
View
Astro Numbers
This project performs digit recognition using deep learning concepts. It can classify an image into 10 classes.We have built a Multilayer perceptron (MLP) model using most popular Google library Tensorflow to recognize handwritten digits.
View
License Plate
Number Plate recognition, also called License Plate recognition using image processing methods is a potential research area in smart cities and the Internet of Things. An exponential increase in the number of vehicles necessitates the use of automated systems to maintain vehicle information for various purposes.
View Repo
InfoViz
The dataset would also include other variables, such as unemployment rates, industrial production levels, and stock prices, depending on the specific economic indicators being analyzed. The visualizations cover a range of economic indicators, including time, unemployment data, consumer price index, price-to-earnings ratio, open, high, low, close, and industrial production index.
View RepoPublications
My research & publicationsGait speed based individual Recognition model using deep 2-D convolutional neural network
Conference : International Conference on Computer Communication and Informatics (ICCCI) | Progress: PublishedLink: 10.1109/ICCCI56745.2023.10128342
Author(s): Abhay Lal; P Nithyakani
Developed a custom 2-D Convolutional Neural Network using thermal infrared night images from CASIA C dataset for gait-based individual recognition, comparison and optimized classification of walking conditions outperformed pre-trained models.
Human Gait Recognition Using Cross View Micro Gait Dataset with Light weight MobileNet
Conference : International Conference on Recent Advances in Electrical, Electronics, Ubiquitous Communication, and Computational Intelligence (RAEEUCCI)Link: 10.1109/RAEEUCCI57140.2023.10134510
Author(s): Haripreeth Dwarakanath Avarur; Abhay Lal; Nithyakani P; Aryan Sinha; Gajulapalli Naga Vyshnavi; Shruthi Kannan
Implemented gait recognition using a lightweight MobileNet model on the Custom-made Cross View Micro Gait Dataset, surpassing existing methods for human identification from cross-view videos.
ALATS: Analysis of localization algorithms in terrestrial surveillance bots
Conference : IEEE International Conference on Electronics, Computing and Communication Technologies (CONECCT)Link: 10.1109/CONECCT57959.2023.10234740
Author(s): Vaishnavi Moorthy; Darshil Shah; Shresth Kapoor; Srinivas Tb; Abhay Lal
Designed and executed a state-of-the-art surveillance system on a Raspberry Pi, harnessing the YOLOv7 model and OpenCV to craft a compact, high-performance autonomous surveillance robot.
Automated screening of hip X-rays for osteoporosis by Singh’s index using machine learning algorithms
Indian Journal of Orthopaedics , Springer | Progress: PublishedLink: 10.1007/s43465-024-01246-9
Author(s): Vijaya Kalavakonda, Sameer Mohamed, Abhay Lal , Sathish Muthu
Developed and evaluated machine learning models for automating osteoporosis diagnosis using the Singh Index from hip radiographs, with the stacked CNN achieving superior accuracy (93.6%) and balanced metrics, making it the most reliable tool for clinical screening.
Unsupervised Machine Learning for Osteoporosis Diagnosis Using Singh Index Clustering on Hip Radiographs
arXiv preprint arXiv:2411.15253 | Progress: PublishedLink: 10.48550/arXiv.2411.15253
Author(s): Vimaladevi Madhivanan, Kalavakonda Vijaya, Abhay Lal, Senthil Rithika, Shamala Karupusamy Subramaniam, Mohamed Sameer
This study addressed the global challenge of osteoporosis by developing a machine learning-based approach to automate Singh Index (SI) identification from hip radiographs, utilizing a custom convolutional neural network for feature extraction and clustering, while highlighting the need for balanced datasets, improved image quality, and the integration of clinical data to enhance diagnostic accuracy.