Himanshu Nimonkar在找工作

Constructed a scalable, high-throughput ML inference pipeline for wildlife monitoring across 4000+ hours of ARU field recordings, integrating CNNs, BiLSTMs, ResNet18, and PANNs with attention-based fusion on multi-resolution spectrograms for temporal-spectral feature modeling, achieving 94% accuracy at 81.5% recall for automated gunshot detection. Systematized model evaluation and experiment management by automating 30+ training and inference runs using PyTorch and Hydra configuration management, reducing evaluation time by 85% and lowering false positives by 70% through rigorous ablation studies and architecture comparisons enabling deployment-ready performance. Designed end-to-end data preprocessing workflows including noise filtering, spectrogram generation, class balancing with SMOTE, and multi-resolution feature extraction, building reproducible training scripts with comprehensive validation dashboards tracking precision, recall, F1, ROC curves, and domain-specific wildlife detection metrics. Containerized the full ML pipeline with Docker for deployment on HPC clusters and cloud environments, improving reproducibility and enabling cross-institutional collaboration between UC Davis and Louisiana State University researchers on bioacoustics and conservation analytics. (Federally Funded Research Collaboration) (Advisor: Prof. Kevin Ringelman (Wildlife, Fish, and Conservation Biology Department of UC Davis)) (Co-Advisor: Prof. McKenzie Fowler (School of Renewable Natural Resources, College of Agriculture, Louisiana State University, LSU)) (Lab: Avian Ecology Lab)

Served as Teaching Assistant for ECS 130: Scientific Computation under Prof. Zhaojun Bai, mentoring 70+ students through numerical analysis and computational methods by leading weekly discussion sections with live coding demonstrations on linear algebra algorithms, iterative methods, sparse matrix computations, and MATLAB/Python implementations while holding regular office hours to provide personalized debugging assistance and academic support. Managed course infrastructure across Piazza, Gradescope, and Canvas platforms, monitoring Q&A forums for timely student support, facilitating peer-to-peer learning, and maintaining consistent communication channels while evaluating student performance by grading midterms, final projects, and weekly assignments with detailed rubrics and constructive feedback. Developed strong communication, mentorship, and organizational skills by adapting teaching approaches to diverse learning styles, fostering an inclusive collaborative environment, coordinating with course staff on curriculum delivery, and maintaining accurate grade records with Excel and Google Sheets to ensure assessment consistency and academic integrity.

Engineered CATF, a production-grade Python ML framework for context-aware multivariate time-series forecasting, implementing a manager–worker architecture with modular system design, unified ETL pipelines, and end-to-end workflows spanning feature engineering, model training, hyperparameter optimization, validation, and deployment across seven attention- and optimal-transport-based models. Designed and implemented MLOps infrastructure including direct CSV-to-tensor ingestion, lazy loading for GPU memory optimization, automated model validation workflows, Optuna-based hyperparameter tuning, multi-dataset CLI support, and CI/CD-ready pipelines to enable reproducible experimentation and scalable model development. Achieved state-of-the-art performance on 5 of 7 benchmark datasets, improving 26 of 35 experimental runs with CATF-TimesNet and CATF-iTransformer reducing MSE by 49.3% and 32.8% respectively through systematic architecture optimization, performance benchmarking, and data transformation efficiency improvements that reduced computation overhead by 9%. Submitted research findings to ACM SIGKDD 2026 Research Track (flagship knowledge discovery and data mining conference), documenting novel context-aware prediction methods and architectural innovations for production ML systems. (Advisor: Prof. Dongyu Liu) (Team Lead: Yueqiao Chen, PhD) (Lab: Visualization and Intelligence Augmentation (VIA) Lab, Computer Science Department of UC Davis)

Architected and deployed a fault-tolerant Stripe API v18 payment pipeline using React Stripe Elements, Node.js verification microservices, and exponential backoff retry logic, implementing comprehensive unit and integration testing with Jest and Supertest to validate failure handling and API contracts, increasing payment success rate to 85% and reducing response time by 30%. Revitalized legacy authentication and communication systems by migrating to secure Node.js/Next.js microservices integrated with Firebase Authentication and AWS SES, centralizing password recovery flows and email preference management with real-time validation, cutting API latency by 35% and boosting recovery completion rate by 80% through optimized backend architecture. Streamlined cloud infrastructure and DevOps workflows by implementing Dockerized CI/CD pipelines on AWS with GitHub Actions, enabling reproducible builds, horizontal scalability planning, and zero-downtime deployments while benchmarking I/O bottlenecks and monitoring system observability through Elasticsearch-backed health dashboards. Designed and optimized REST API contracts between React/TypeScript frontend and Node.js/Next.js backend services using Tailwind CSS for responsive UI, improving end-to-end application responsiveness by 40% through systematic API performance tuning and backend optimization.

Engineered an Android security testing application using Kotlin and Android SDK to identify permission vulnerabilities in system functions including location services (Google Maps API), camera, microphone, SMS, and media access, implementing automated security audits that reduced permission-related incidents by 42%. Developed and deployed a real-time CVE/CWE vulnerability tracking system with user-configurable watchlists and automated email alerting via SendGrid API for proactive threat monitoring, while optimizing CI/CD workflows using GitHub Actions and Docker, cutting deployment time by 35%. Built a CLI-based red teaming automation and OSINT toolkit using Python and Bash scripting for penetration testing and threat modeling, uncovering 37 new attack vectors through systematic security assessments and improving malware defense capabilities by 29%. Architected DarkwebGPT, a threat intelligence platform leveraging the DarkBERT API and transformer-based NLP for automated dark web content analysis, and developed a LangChain-based legal assistant using Hugging Face transformers on IPC datasets with RAG architecture, increasing user engagement by 25%. Led AI-driven cybersecurity research initiatives including chatbot development, workflow automation, and the Cybersafe Mumbai public awareness campaign, while investigating 10+ cybercrime cases spanning social media breaches, financial fraud, data leaks, malware attacks, and crypto scams. Built and maintained a comprehensive AI tools database presented to the Office of the Governor of Maharashtra, and developed an AI-powered content management automation system integrating multiple AI APIs to streamline creative workflows and accelerate digital content production.

Conducted applied human-centered AI and UX research across education, mental health, pet welfare, and student productivity domains, leading end-to-end studies including literature review, TA workflow analysis, stakeholder interviews, and large-scale online and in-person surveys to identify real user pain points, while translating qualitative and quantitative findings into system requirements and UI prototypes evaluated through comparative studies, preference analysis, and adoption metrics. Published 4 peer-reviewed papers (2 IEEE, 2 Springer) on domain-specific application design: "Streamlining Educational Assessment" (ICAST 2023, IEEE) designing AI-driven grading workflows based on TA surveys; "Enhancing Student Welfare" (ASCIS 2023, Springer) achieving 87.4% student preference for university mental health counseling UI; "Enhancing Pet Welfare" (ICTACS 2024, IEEE) developing GPS and health tracking interfaces; and "UniExplorer" (ICTACS 2024, IEEE) creating opportunity management platforms for internships, research, and campus activities, validated through user testing and feedback. Collaborated with Somaiya School of Design under Prof. Ashwini Dalvi and Prof. Irfan Siddavatam, evaluating usability, accessibility, and task efficiency across multiple UI prototypes to inform evidence-based design decisions for production deployment.

Designed and implemented a large-scale NLP modeling pipeline to predict integer app ratings (1–5 stars) from user reviews, scraping and preprocessing 597K+ iOS App Store reviews across 97 apps with extensive text normalization, lemmatization, stopword removal, contraction expansion, and rigorous train/validation/test splits. Developed and systematically compared sequence-based and transformer architectures for multi-class classification and regression, implementing RNN, LSTM, bidirectional LSTM, and ELECTRA-Mini models with optimized embeddings, dropout regularization, batch/layer normalization, AdamW optimization, warmup scheduling, and Optuna-based hyperparameter search. Led comprehensive architecture exploration across hidden layer dimensions, dropout rates, weight decay parameters, and StepLR learning rate scheduling using PyTorch and Hugging Face Transformers, identifying stability-generalization trade-offs and implementing data augmentation strategies to mitigate overfitting on noisy user-generated text. Built an optimized bidirectional LSTM regression model with layer normalization and synonym-based augmentation achieving 53% validation accuracy while remaining computationally lightweight (~34 MB, 505M FLOPs), validating performance through confusion matrices and learning curve diagnostics showing 80%+ predictions within ±1 rating error. Co-authored "User Reception is Everything: Using a Neural Network to Predict iOS App Ratings" published in Springer's Data Science and Applications (2024), documenting transformer-based rating prediction methodology and proposing ELECTRA-based systems achieving 72.96% test accuracy with 0.705 F1 score for production deployment. (Advisor: Prof. Ashwini Dalvi, Prof. Irfan Siddavatam)

Engineered a geospatial analytics pipeline using ArcGIS Pro and Python to map and analyze agricultural land for Somaiya Agro Trust, performing spatial data processing and feature extraction for soil composition, terrain topology, and crop suitability assessment across multiple village clusters. Developed machine learning models using scikit-learn and pandas for predictive analysis of farming conditions and agricultural viability, implementing scalable data pipelines to enable village-level land assessment and expand analysis to larger rural regions with automated batch processing. Integrated GIS-based spatial features with ML workflows by converting shapefiles and raster data into structured datasets, building ETL pipelines to merge geospatial attributes with predictive models for data-driven agricultural planning and land optimization recommendations. Collaborated with domain experts and agricultural scientists to refine geospatial datasets, validate ground-truth field conditions through iterative feedback loops, and prepare production-ready deployment architecture for large-scale village assessment tools. (Advisor: Prof. Ujwala Bhangale, Prof. Nilkamal Prashant More)

Designed and implemented a computer-aided diagnostic pipeline for early knee osteoarthritis detection from X-ray imaging, leveraging transfer learning with DenseNet-201 initialized on ImageNet weights and optimized for low-data clinical settings with limited labeled medical imagery. Led end-to-end ML experimentation including dataset curation of 5,700+ labeled X-ray images from public medical repositories, class imbalance mitigation via RandomOverSampler, TensorFlow ImageDataGenerator-based augmentation (rotation, shifts, zoom, shear), normalization, and preprocessing workflows tailored for medical imaging. Developed and trained a binary classification model using Keras and TensorFlow with early stopping and exponential learning rate scheduling, evaluating performance through confusion matrices and probability-based outputs to achieve 82.48% test accuracy, 97.62% precision, and 77.36% F1 score critical for clinical decision support. Authored and presented "A Transfer Learning Approach for Classification of Knee Osteoarthritis" at IEEE ICEEICT 2023, receiving Best Paper Award in technical session CS 03, while conducting comprehensive literature review comparing traditional ML models and 3D CNN approaches to validate DenseNet-based transfer learning as a computationally efficient solution. Designed system architecture for integrating AI-based OA screening into clinical workflows including data ingestion, preprocessing pipelines, model inference APIs, and result visualization dashboards for radiologist decision support, analyzing precision–recall trade-offs essential for medical AI deployment. (Advisor: Prof. Ninad Mehendale)

Built a scalable Python-based network graph visualization system processing 116,000+ web crawling entries, implementing data ingestion pipelines using pandas for ETL operations including cleaning, deduplication, normalization, and attribute enrichment with structured CSV and Excel exports. Developed an interactive, production-ready network visualization web application using NetworkX for graph computation and Bokeh for dynamic rendering, implementing advanced user interactions including scroll zoom, box selection, pan navigation, hover highlighting, node linkage visualization, graph reset, and PNG export functionality. Engineered efficient graph rendering algorithms by optimizing node and edge data loading through direct CSV ingestion, mapping node IDs to origin IDs for accurate relationship construction, and implementing lazy loading techniques to handle large-scale datasets under performance constraints. Designed a two-page responsive web interface with custom HTML/CSS styling, creating an introductory landing page (index.html) and embedding the NetworkX visualization via iframe with smooth page transitions triggered by JavaScript event handlers for enhanced user experience. Implemented real-time visual analytics features including dynamic node highlighting on hover, edge relationship tracing, and aesthetic graph layout algorithms to enable intuitive network analysis and pattern discovery for non-technical stakeholders. (Advisor: Prof. Ashwini Dalvi, Prof. Irfan Siddavatam)

Designed and developed a 10-page production WordPress website from scratch following modern UI/UX principles, implementing responsive design with custom CSS, mobile-first layouts, and cross-browser compatibility testing to ensure optimal user experience across devices. Integrated four interactive user forms using Contact Form 7 and custom JavaScript validation, embedded Google Maps API with custom markers for office location display, and implemented form submission handling with email notifications and database logging for lead management. Architected and deployed a secure payment processing system by integrating Stripe API for debit/credit card donations, implementing webhook handlers for payment verification, transaction logging, and automated receipt generation while ensuring PCI DSS compliance and secure data handling. Managed full domain deployment workflow including domain purchase from GoDaddy, DNS configuration and propagation, domain transfer to WordPress hosting, SSL certificate installation for HTTPS encryption, and server configuration for production-ready website launch with 99.9% uptime. Optimized website performance through caching plugins, image compression, minified CSS/JS assets, and CDN integration, achieving sub-2-second page load times and implementing SEO best practices including meta tags, structured data, and XML sitemaps.

Developed computer vision models using YOLO v4-tiny and YOLO v5 architectures with PyTorch for real-time underwater object detection, training custom datasets to identify mission-critical targets including gates, flares, markers, and marine debris with 85%+ mAP accuracy, and built ROS Melodic software modules in Python/C++ for autonomous navigation, sensor fusion, and mission planning with Gazebo simulation environments enabling rigorous testing before hardware deployment. Contributed to research on path planning algorithms and image processing pipelines, co-authoring technical papers on underwater navigation systems, implementing A* and RRT* pathfinding algorithms, and conducting systematic experimentation to optimize model performance under varying water conditions while designing frontend components for the team's website using HTML, CSS, and JavaScript. Led a 22-member interdisciplinary engineering team as Vice Captain, coordinating project management including budget allocation ($25K+), logistics planning, sprint scheduling, and timeline management using Trello and Gantt charts, while serving as liaison between technical sub-teams, faculty mentors, Students' Council, and college administration through weekly standup meetings and project documentation. Delivered comprehensive technical training including hands-on ROS Melodic workshops for 22 students covering publishers, subscribers, services, and TF transforms, taught OpenCV image processing fundamentals to 26 participants with Python coding exercises, and supervised technical documentation for AUV Proton while mentoring sub-teams across software, electrical, mechanical, and simulation domains.

Courses Taken=> Fall 2024:- ECS 265: Distributed Database Systems (Prof. Mohammad Sadoghi) ECS 271: Machine Learning & Discovery (Prof. Hamed Pirsiavash) Winter 2025:- ECS 260: Software Engineering (Prof. Vladimir Filkov) STA 220: Data & Web Technologies for Data Analysis (Prof. Peter Kramlinger) Spring 2025:- ECS 289L: AI in Health (Prof. Vladimir Filkov) ECS 222A: Design & Analysis of Algorithms (Prof. Yelena Frid) Fall 2025:- STA 221: Big Data & High Performance Statistical Computing (Prof. Nicolai Amann) Winter 2026:- ECS 220: Theory of Computation (Prof. David Doty)

B.Tech in IT: Applied Maths I & II Programming in C Engineering Drawing Elements of Electrical & Electronics Engineering Engineering Chemistry Python programming Engineering Physics Engineering Mechanics Environment & Technology Data Structures Advanced Python Programming Discrete & Applied Mathematics Database Management Systems Data Communication & Networking Digital Systems Developing Web Applications with Python & Flask Probability, Statistics & Optimization Techniques Information Theory & Coding Analysis of Algorithms Advanced Databases Competitive Programming Theory of Computation Operating System Information & Network Security Web Programming (Front-End & Server Side) UI Programming Object-Oriented Software Engineering Modeling & Simulation Cloud Computing Vulnerability Analysis & Penetration Testing Software Testing & Quality Assurance Digital Marketing DevOps Honors in AI: Data Science Intro to Artificial Intelligence Machine Learning Deep Learning Natural Language Processing

- Completed Electronics Maintenance Vocational Course.