K
Kishorkumar Devasenapathy
About
Detail
Gujarat, India
Résumé
Jobs
verified_user
0% verified
Software EngineerBrexFeb 2025 - Current (1 year 5 months)Worked on Brex Dashboard Redesign for Financial Teams and collaborated with product, design, and research teams during agile sprints. Led requirement gathering sessions and improved dashboard usability and financial data clarity for end users by 40%. Built reusable frontend components using React and MUI. Implemented Redux Toolkit for state management and A/B testing for key flows which improved user interaction and feature engagement by 28%. Optimized database queries using AWS RDS and added caching to speed up responses. Implemented security layers with AWS KMS and IAM which reduced data access vulnerabilities and compliance risks by 60%. Developed backend features in Spring Boot including budget analytics, approval workflows, and transac- WSoftware EngineerWN4 Smart Systems LabMay 2024 - Jan 2025 (9 months)Developed and deployed ETL pipelines for meteorological and wireless KPI datasets, processing over 190,000 data points from 24 months to support accurate and scalable time-series analysis in CBRS wireless networks. Designed and implemented an artificial neural network for signal strength prediction, achieving a low mean absolute error of 1.83 dB (2.81%), significantly improving forecasting accuracy across multiple urban and rural regions. Engineered domain-specific features by integrating environmental and network telemetry, enabling more precise modeling of wireless signal behaviors over time, and facilitating robust data ingestion and transformation across distributed systems. Containerized and orchestrated machine learning training jobs
- WBackend Web DeveloperWipro ltd.Aug 2021 - Jul 2023 (2 years)Developed and delivered 18 robust backend API features for FedEx account administration, leveraging Java, Spring, REST, JSP, JDBC, and ORM within an MVC architecture to enhance system functionality and reliability. Spearheaded the integration of two-factor authentication (2FA) across multiple system teams, significantly improving security protocols and safeguarding sensitive account access in compliance with industry best practices. Identified and resolved two critical zero-day vulnerabilities impacting user password security, ensuring immediate risk mitigation and strengthening overall system resilience against cyber threats. Coordinated cross-functional collaboration among database, tracking API, and frontend teams to streamline developme
- MResearch AssistantMarmot LabDec 2020 - Jul 2021 (8 months)Developed autonomous multi-drone navigation using distributed Deep Reinforcement Learning, achieving 87% efficiency versus centralized systems, enabling robust decision-making in dynamic environments with real-world constraints like limited communication and partial observability. Designed a scalable, adaptive architecture for multi-agent coordination, outperforming fixed-policy algorithms across varying environment sizes by optimizing exploration, responsiveness, and fault tolerance in decentralized drone systems using learned behaviors instead of hardcoded rules. Implemented Asynchronous Advantage Actor-Critic (A3C) with motion and sensing constraints, allowing real-time policy learning and control under asynchronous settings, supporting
Education
verified_user
0% verified
- UMaster of Science (M.S.) in Computer Science and EngineeringUniversity at Buffalo - SUNY, USAAug 2023 - Jan 2025 (1 year 6 months)
- NBachelor of Engineering in Electrical and Electronics EngineeringNational Institute of Technology- Tamil Nadu, IndiaAug 2017 - May 2021 (3 years 10 months)
Projects (professional or personal)
verified_user
0% verified
- OOperational Maintenance of VehiclesCreated an 18-class CNN with encoder-decoder architecture for defect detection in vehicle components, achieving 82% classification accuracy. Led a 6-member team in building and deploying a TFLite model on Android with 15 fps inference speed, ensuring real-time defect recognition.
- MModel Quantization ResearchOptimized RelPose model for mobile inference using TensorFlow Lite and NVIDIA AMP, reducing model size and latency while enhancing model serving practices. Executed parallelized training on dual RTX 3080TI GPUs across 1.4 TB of image data, demonstrating efficiency in large-scale model training pipelines.
- HHealthcare Provider Matching SystemBuilt a LangFlow-based RAG agent using LLama LLMs, achieving 89% accuracy in multilingual healthcare provider matching with prompt engineering and validation. Designed a low-tech literacy-friendly StreamLit interface with built-in translation using NLP techniques, enabling 82% accurate multilingual support.
- WWindspeed Prediction for Solar Energy ForecastingDeveloped a stateful LSTM RNN to forecast hourly wind speed and solar radiation, achieving a 4.75% average error using time-series models in TensorFlow and PyTorch. Preprocessed 500K data points across 4 years by handling sparse features and multi-collinearity in a 38-feature weather dataset to improve model stability.