• Designed, developed, and maintained backend software systems using Java and Spring Boot to support scalable multi-service applications.

• Improved CI/CD infrastructure by refactoring GitHub Actions workflows, reducing build time by 30% and improving deployment reliability.

• Developed automated database migration pipelines using Flyway and PostgreSQL to maintain schema consistency across distributed environments.

• Integrated machine learning services by implementing OpenAI Chat Completion APIs for sentiment analysis and automated content moderation.

• Deployed AI-driven spam detection systems using Azure AI models, reducing inappropriate content by 25% with feature-flag based rollout strategies.

• Collaborated with cross-functional teams to design, test, deploy, and monitor software services in cloud environments.

• Proposed HALP: single-pass probes on VLM internals to predict hallucination risk before decoding, replacing post generation detection.

• Trained probes on ∼ 250k-sample, six-benchmark VQA; late transformer layers achieved AUROC ∼ 0.93 on Gemma-3, Llama, LLaVA, Qwen.

• Built a 3 Agent workflow to synthesize image–text pairs: VLM-guided image generator → QA generator → validator, with batch orchestration.

• Proposed a novel hybrid NER architecture to address annotation inaccuracies and generative hallucinations in biomedical entity recognition.

• Developed a multi-stage fusion architecture combining BERT-based classifier with RAG LLM as a judge for entity-level factual grounding.

• Designed and implemented full-stack web applications using React.js, Node.js, and cloud services to support scalable platform functionality.

• Developed modular UI components and backend services that automated operational workflows, reducing manual workload from 40 hours/week to minutes.

• Built analytics dashboards using Chart.js and user behavioral data to generate insights that increased revenue by 30%.

• Implemented automation pipelines and scheduled scripts for referral code generation, increasing new user signups by 23%.

• Collaborated with engineering teams to deploy scalable web services on AWS EC2 and Firebase cloud infrastructure.

• Developed computer vision models using CNNs with transfer learning (VGG16) for weather prediction applications.

• Trained and evaluated deep learning models achieving 94% prediction accuracy.

• Implemented model training pipelines using TensorFlow and Python for scalable ML experimentation.