Vijay Murugan Appavu Sivaprakasam
Backend & AI Engineer
Building intelligent, scalable, and robust systems. Specialized in backend development and applying AI to solve complex problems.
About Me
I am a passionate backend and AI engineer with a knack for designing and building scalable, high-performance applications. With a strong foundation in software architecture and distributed systems, I specialize in creating the backbone for modern web services. My journey into artificial intelligence has equipped me with the skills to integrate machine learning models, build intelligent features, and leverage data to create smarter, more efficient solutions. I thrive in collaborative environments and am always eager to tackle new challenges and learn new technologies.
Skills
Backend
AI & Machine Learning
Databases
DevOps & Cloud
Frontend
Monitoring & Logging
Soft Skills
Experience
My professional journey in backend engineering and software development.
Key Achievements:
- •Developed a payment feature for subscription-based services using Spring Boot and integrated with third-party payment gateways
- •Worked on the development of multiple automations to settle payments and refunds
- •Migrated over 2M+ records from MySQL to DynamoDB using multithreaded scripts
- •Orchestrated payment APIs to consolidate transactions across multiple services
- •Provided on-call support and live debugging for critical, customer-facing transaction issues
- •Designed an address verification mechanism for Indian customers that reduced fraudulent activities and manual review processes by 10% while maintaining 98% accuracy
- •Implemented AI-based user retention strategies that improved customer retention rates
- •Mentored 2 junior engineers and led code review processes
Technologies:
Key Achievements:
- •Spearheaded a Fast API microservice to build an address handling system for users that supported multiple address
- •Added logs and error handling in the existing codebase to improve debugging and error tracing
- •Monitored logs and alerting using Splunk and Sumo Logic
Technologies:
My Work
A selection of my projects, showcasing my skills in backend engineering and artificial intelligence.
Implemented diffusion based Generative AI adversarial attacks on images and through loss function optimization by modifying cross-entropy loss to Carlini & Wagner loss, improving the attack transferability by 39%. Researched about various image immunization techniques including PhotoGuard, AdvPaint, and DiffusionGuard, evaluated PhotoGuard on a small subset of images, achieving a success rate of 96.67% in preventing attacks
Evaluated a visual chain-of-thought framework on the Path-VQA dataset containing 6k+ pathology Q&A pairs, focusing on yes/no accuracy and qualitative reasoning quality. Benchmarked GPT-4o, GPT-3.5-mini, Gemini-2.5-Flash, and BLIP-VQA models on medical visual reasoning, analyzing performance gaps and model strengths across tasks and reported a 10% improvement over baseline reasoning performance
Engineered an ML-based Android + IoT system for grocery classification and weight detection, achieving 95%+ accuracy, 98% precision, and 90%+ real-world accuracy using custom datasets and ensemble models (EfficientNetV2, MobileNetV2) Published findings at ICSMDI 2023 and secured an Indian Patent (#555691), showcasing the application of ML research to practical retail challenges by reducing checkout wait times and enabling scalable deployment
Optimized self-supervised CLIP training by testing against multiple optimizers including Adam, AdamP, and AdamW and various loss functions SogCLR and iSogCLR achieving best zero-shot top 1 accuracy of 24.45% using AdamW with SogCLR loss. Added Weights & Bias for checkpoint management to reduce the training time by 87%
Engineered a full-stack bill splitting application using FastAPI, React, and MongoDB, building 12+ API endpoints and 8+ interactive UI components for multi-user expense management, processing split transactions with automated balance tracking
Engineered a scalable React.js application with Firebase authentication serving 250+ users at 99% reliability, integrating Google Favicon API, MongoDB storage, and canvas-based development tools to build flow-charts with company logos.
Built a personalized movie recommendation system by training a hybrid ML model combining K-means and cosine similarity. Deployed with Fast API backend, TypeScript frontend, and Firebase authentication to deliver over 90% accurate suggestions.
Developed an intelligent PDF content enhancer using advanced NLP techniques to automatically summarize and extract key information from documents. Integrated with a user-friendly interface built on React and FastAPI.
Get in Touch
Have a question or want to work together? Leave a message.