AI/ML
AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
I build foundational AI/ML systems that move research from prototype to production at real-world scale. My work spans speech, language, and multimodal learning serving healthcare, biopharma, and consumer/enterprise platforms. My focus is on data, modeling, and evaluation, alongside the infrastructure and systems optimization that make ML measurable, scalable, and dependable in deployment.
Andhra University
Bachelor of Engineering - BE, Electronics and Communications Engineering
N/A – Present
Arizona State University
Master of Science (M.S.), Computer Engineering
N/A – Present
Apple
Machine Learning Scientist
March 1, 2024 – Present
Microsoft
Applied Scientist
June 1, 2022 – June 1, 2023
Abridge
AI Scientist-Engineer
June 1, 2018 – June 1, 2022
The Luminosity Lab
Graduate Student Researcher
January 1, 2016 – January 1, 2018
Arizona State University
Teaching Assistant
January 1, 2016 – January 1, 2016
Text Independent Speaker Identification
August 1, 2019 – September 1, 2019
Implementation of speaker verification using Google's GE2E loss function, with added noise corruption testing at different SNR levels.
Sentence Similarity Evaluation
January 1, 2018 – March 1, 2018
Built two parallel models: Deep Siamese LSTM w/word2vec embeddings (Keras), and traditional ML (LightGBM/XGBoost) using 7 linguistic features (Jaccard/TFIDF/GloVe/para2vec/n-grams). Ensembled for 93% accuracy. Leveraged Keras for rapid prototyping.
Visual Search
September 1, 2017 – October 1, 2017
An efficient image retrieval pipeline combining: - Fast image retrieval using Ball Trees on AlexNet features reduced to 80 PCs - Result diversification through iterative k-means clustering (k selected via elbow method) - Experimentation with Siamese Networks to improve feature discrimination
Fraud Detection in Financial Systems
September 1, 2017 – September 1, 2017
• Review of various machine learning techniques such as k-Means, kNN, Decision Trees and Bayesian networks applied to fraud detection. • Implemented an Autoencoder and Principal Component Analysis features on Credit Card Fraud dataset to detect anomaly distribution.
Clothing Attribute Recognition
July 1, 2017 – August 1, 2017
Transfer learning image classifier built on VGG MaxPool-4 features with a lightweight CNN head (2 layers + BatchNorm + GAP), trained with standard augmentations and evaluated using comprehensive F1 metrics.
Two Stream LSTM for Human Activity Recognition
May 1, 2017 – July 1, 2017
Developed video classification model using Brox Optical flow and ResNet-101 features, combined with LSTM for temporal modeling. Enhanced with stacked flow frames and data augmentation for robust spatiotemporal understanding.
Kaggle Competetion: Mercedes-Benz Greener Manufacturing
May 1, 2017 – May 1, 2017
Engineered hybrid model combining dimensionality reduction (PCA, ICA, Random Projections, Autoencoders) with ensemble of Gradient Boosting and LassoCV, weighted with XGBoost (75/25 split). Achieved top 1% leaderboard ranking.
Automatic Image Colorization
February 1, 2017 – April 1, 2017
Developed cross-platform image colorization system using VGG-based Deep Learning model with hypercolumn feature extraction. Deployed solution as scalable REST API on Google App Engine, with companion Android and web applications.
RBM based Autoencoder for MNIST dataset
January 1, 2017 – January 1, 2017
Weights for Autoencoder are obtained by pre-training with RBM. Used Contrastive Divergence algorithm for training RBM. Observed that these pre-trained weights have greatly improved the convergence of the Autoencoder. Design and Inspiration: G. E. Hinton and R. R. Salakhutdinov's Reducing the Dimensionality of Data with Neural Networks
Geospatial Data Processing Operations using Apache Spark and HDFS
January 1, 2017 – April 1, 2017
Built scalable geospatial system (Spark/HDFS) optimizing range/kNN/join queries. Benchmarked performance across varying data loads & cluster sizes. Identified NYC taxi hotspots (n=50, p<0.01) on 4-node deployment. Demo: https://www.youtube.com/watch?v=CIXGdCKWQp8
CloudPi Web Application using Amazon Web Services
January 1, 2017 – February 1, 2017
Develoepd FFT-driven Pi calculator with Multi-tier AWS architecture - Web Tier: S3 input/output handling via PHP SDK - Queue: SQS FIFO for request management - App Tier: Dynamic auto-scaling (CPU-based) Included CLI tooling for instance management. Optimized for cost/performance balance.
To do List, a smart task management application for Android
September 1, 2016 – December 1, 2016
Designed a Smart Task Manager app with core features: - Data: SQLite + Google Drive sync - UX: Image tasks, priority levels, adaptive brightness - ML: Frequency-based task prediction (uses 3-month rolling history), Learns from user patterns - helps predict what the user might need to do next.
A Recursive Neural Network for Factoid Question Answering over Paragraphs
September 1, 2016 – December 1, 2016
Implemented a deep recursive neural network model for accurately answering paragraph-length factoid questions. This innovative model demonstrates the ability to reason over question text, even when it contains limited clues indicative of the answer. Applied this cutting-edge model to a dataset comprising questions from the quiz bowl trivia competition, showcasing its performance in handling complex and nuanced queries.
8 bit Modulo Adder Design
January 1, 2016 – April 1, 2016
- High-performance 8-bit adder implemented in 32nm technology, achieving 1.919GHz operation and 74.35 pJ.ps EDP. - Used mirror topology with TSPC registers for synchronized pipelined operation. - Design and verification completed through industry tools: Virtuoso, Hspice, Hercules, and StarRC.
CyanogenMod based Android Custom ROM for OnePlus and Yu flagship devices
February 1, 2015 – March 1, 2016
- Established CyanogenMod development environment on ArchLinux, syncing repositories and device trees. - Cross-compiled ARM kernel using Linaro/Uber toolchains, implementing CPU overclocking features. - Integrated prebuilt user applications and kernel into the final ROM build.
Bank Locker Security System
June 1, 2014 – June 1, 2014
- Developed security alarm system using AT89S52 microcontroller integrated with GSM module, keypad interface, and motor control. - Programmed core functionality in Embedded C using Keil compiler, complete peripheral interfacing with MAX232 and L293D motor driver.
VHDL Modeling of Convolutional Interleaver-Deinterleaver for Efficient FPGA Implementation
January 1, 2014 – January 1, 2014
- Designed efficient convolutional interleaver model on FPGA using VHDL, implementing embedded shift registers for burst error handling. - The architecture achieved 81% reduction in FPGA resource utilization through optimized incremental shift register implementation.
Cultural Fit Analysis
The candidate's project portfolio demonstrates a strong inclination towards innovation and practical application of advanced technologies, which aligns well with a dynamic, research-oriented ML environment. Their experience at companies like Apple and Microsoft, coupled with contributions to open-source initiatives (CyanogenMod), suggests a collaborative and impact-driven mindset. The breadth of projects, from low-level hardware to high-level ML applications, indicates a versatile individual who can contribute across different facets of an engineering team.
Soft Skills & Operational Fit
The candidate's project descriptions indicate a proactive and innovative approach to problem-solving, often involving novel architectures and optimization. Experience as a Graduate Student Researcher and Teaching Assistant suggests an ability to learn, teach, and collaborate. The diversity of projects, from embedded systems to cloud applications, indicates adaptability and a broad technical curiosity.