Final Year engineering student

Ashutosh Kumar  • Peerlist

• Accelerated API performance by leveraging Redis for caching, achieving a 75% reduction in latency, and engineered a Redis-based rate limiter in Express.js to reduce server load and enhance security. • Architected and deployed a scalable microservices infrastructure using React.js, TailwindCSS, and Recoil.js, improving system reliability and scalability. • Leveraged TypeScript to enhance code quality and maintainability, ensuring type safety and reducing runtime errors across the application. • Optimized data management with PrismaORM and PostgreSQL, ensuring data integrity and improving query performance.

brevity

This project focuses on building a music genre classification system using deep learning. It utilizes MFCC (Mel Frequency Cepstral Coefficients) as audio features and combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks to learn spatial and temporal features from audio data. The model is trained using K-Fold Cross-Validation, and a frontend allows users to predict the genre of an audio file.

Music Genre Classification using CNN + LSTM

• Built a real-time face authentication system using React.js, Node.js/Express, and face-api.js, integrating SSD MobileNet V1 for face detection and ResNet-34 for facial recognition. • Implemented facial embedding-based registration and authentication with PostgreSQL and Prisma ORM for secure, structured data storage. • Enabled real-time webcam processing using WebRTC (getUserMedia) and displayed live detection overlays with Tailwind CSS for UI. • Containerized the app using Docker, and utilized Socket.io for real-time authentication and server-client event handling.

Realtime Face Based Authentication With Age &

• Deployed distributed tracing using OpenTelemetry and Jaeger to monitor and analyze microservices performance. • Integrated OpenTelemetry tracing with a gRPC-based service, capturing request propagation across distributed systems. • Configured Jaeger as a trace collector via OTLP and enhanced span reporting using batch processing for efficiency. • Developed comprehensive tracing solutions with OpenTelemetry integrated into gRPC systems, identifying critical data flows that accelerated debugging by 30 hours per month.

Cloud-Native Tracing with OpenTelemetry & Jae

• Streamlined video processing by pulling events from Kafka and streaming videos directly from S3 to FFmpeg without local storage. • Orchestrated transcoding in isolated Docker containers, generating multiple resolutions (480p, 720p, 1080p), and uploading results to S3. • Devised a seamless integration between AWS Lambda and Kafka that triggered real-time events on new uploads; • Improved data processing workflows while reducing manual intervention on critical upload notifications by nearly two hours daily. • Enhanced resource utilization and ensured fault tolerance with efficient message offset management and monitoring. • Integrated Prometheus and Grafana for real-time monitoring, visualizing ECS task performance, and tracking system metrics.