AI Researcher & Full-Stack Developer
I work on probabilistic modeling and machine learning systems that quantify uncertainty. My research bridges statistical foundations with modern deep learning, focusing on methods that are both theoretically grounded and practically scalable. Primary interests include Bayesian inference, variational methods, generative modeling, and robustness under distribution shift.
Currently developing research in uncertainty quantification for neural networks, scalable approximate inference, and generative models for structured data. I approach problems through the lens of information theory and statistical mechanics—seeking to understand fundamental limits and tradeoffs in learning systems.
I combine rigorous mathematical analysis with empirical investigation. Every research direction is grounded in specific technical problems, not aspirational goals. My approach privileges understanding over adoption—diving deep into fundamental questions about learning systems rather than chasing trends.
Professional progression from systems engineering toward research-focused machine learning. Current focus reflects intentional trajectory toward PhD-level research in probabilistic AI and statistical foundations.
Amrita Vishwa Vidyapeetham
Research Value
Integrates classical statistical inference with modern deep learning architectures. Emphasizes uncertainty-aware modeling for interpretable, deployable AI systems grounded in statistical rigor.
Rinex Technologies Pvt. Ltd. & Hipla Technologies
Research Value
Applied security principles to system-level threat modeling and mitigation—applicable to robust, adversarial-resistant AI systems.
Chegg India Pvt. Ltd.
Research Value
Deepened statistical foundations and technical communication—essential for research publication and knowledge transfer.
Career Trajectory: Systematic progression from full-stack systems development toward research-focused machine learning and probabilistic modeling. Early experience in security architecture and systems design established foundational understanding of computational constraints, scalability requirements, and principled engineering practices. Current focus on healthcare AI and uncertainty quantification represents intentional transition toward PhD-level research. Production experience directly informs research priorities: deployed systems must be reproducible, computationally efficient, and rigorously validated.
Deep learning system for automated wound detection, segmentation, and classification with clinical decision support.
Technologies
Real-time wound analysis with treatment recommendations for clinical and remote settings
Modular robotic pipeline integrating perception, planning, and actuation with real-time object detection and feedback control.
Technologies
End-to-end system combining perception and actuation with sub-100ms latency
Reinforcement learning model optimizing traffic signal timings with custom simulation environment and policy evaluation.
Technologies
15-20% improvement in traffic flow over traditional fixed-timing systems
Secure login system with multi-layer defense against SQL injection, XSS, and unauthorized access.
Technologies
Defense-in-depth architecture with proven security audit compliance
Secure social networking platform for Amrita University with encrypted REST APIs and optimized mobile architecture.
Technologies
Modular architecture with <50MB memory footprint on resource-constrained devices
Full data pipeline for trip data cleaning, analysis, and visualization with trend discovery and pattern identification.
Technologies
Data-driven insights into pickup patterns, fares, and passenger behavior
Low-level system design implementing complete computing stack from HDL gates through virtual machine architecture.
Technologies
Foundational understanding of computing abstraction layers and systems design
Developing scalable approximate inference methods combining variational techniques with modern hardware accelerators. Focus on structured variational families and amortized inference for high-dimensional distributions.
Quantifying aleatoric and epistemic uncertainty in neural networks without sacrificing prediction accuracy. Connecting Bayesian principles to modern deep learning practices through principled uncertainty estimation.
Building generative models that respect domain structure and constraints. Bridging discrete and continuous representations for scientific data generation.
Investigating fundamental limits of learning algorithms through information theory and statistical mechanics. Understanding when and why modern deep learning generalizes despite high dimensionality.
Current work emphasizes the intersection of uncertainty quantification and scalable inference. Pursuing PhD to deepen investigation into statistical foundations and develop practical algorithms addressing the gap between theoretical guarantees and empirical performance.
Methodology
Developed generalized additive models and Bayesian hierarchical structures to capture population-level and patient-specific wound dynamics. Implemented variational inference for scalable posterior estimation.
Methodology
Implemented Deep Q-Networks with experience replay and target networks. Trained on simulated traffic scenarios with reward shaping for average wait time minimization.
Methodology
Ensemble of VGG-19 and InceptionV3 with temperature scaling for calibration. Implemented conformal prediction framework for distribution-free uncertainty quantification.
Methodology
Implemented SHA-256 password hashing with salt. Used parameterized queries and context-aware output encoding. Designed defense layers: input validation, prepared statements, output encoding.
Methodology
Built perception pipeline using OpenCV for real-time detection. Developed motion planning using Gazebo simulations. Integrated ROS2 nodes for sensor fusion and command execution.
Methodology
Implemented AES-256 encryption for data at rest and transit. Used Android Jetpack for modular architecture. Optimized memory footprint through JVM profiling.
Peer-reviewed publications and research contributions advancing knowledge in AI, healthcare, and environmental science.
Research Philosophy: Published work reflects rigorous methodology and honest reporting. Each publication represents months of experimental validation, statistical analysis, and careful interpretation. I prioritize reproducibility—all work is built on publicly available datasets and transparent methodology.
Beyond research theory, I build systems. This requires understanding practical constraints—computational budgets, numerical stability, data quality, scalability—and translating research ideas into functional implementations that perform reliably in practice.
Design and implement end-to-end machine learning pipelines from data ingestion through evaluation. Experience with distributed training, efficient inference, and production-level systems optimization.
Build inference systems using JAX and PyTorch. Implement custom variational objectives, MCMC kernels, and amortized inference networks. Focus on numerical stability and computational efficiency.
Rigorous experimental design including hyperparameter tuning via Bayesian optimization, statistical significance testing, and reproducible evaluation protocols. Track experiments systematically for rigorous scientific comparison.
Construct robust data pipelines for preprocessing, augmentation, and efficient batching. Experience with distributed data processing and handling real-world data quality issues.
Proficient in implementing and modifying neural architectures. Understanding of computational graphs, automatic differentiation, and practical optimization considerations.
Version control for code and experiments. Clear documentation of methodology, dependencies, and evaluation protocols. Focus on enabling others to understand and reproduce work.
How can we design variational families that maintain expressiveness while remaining tractable at extreme scale?
What is the fundamental relationship between uncertainty magnitude and model generalization under distribution shift?
Can we develop unified frameworks for uncertainty quantification that work across supervised, unsupervised, and RL settings?
How do information-theoretic bounds on learning translate to practical algorithm design?
What architectural properties of neural networks enable calibrated uncertainty estimates without post-hoc methods?
How can we incorporate domain knowledge as constraints in generative models while maintaining tractability?
These questions reflect current research priorities. They emerge from technical gaps encountered in recent work and represent directions where progress has potential for significant impact on both theory and practice.