* Built LangGraph based Dynamic Agent workflows orchestrating PyTorch inference for fine-tuned models via JS Rhino async pipelines, processing emails, PDFs, and web data at 10k+/hour.

* Architected RAG pipelines with pgvector-based similarity search, abstracted retrieval and generation layers to remain vector-database and LLM agnostic, optimizing chunking, embeddings, and reranking.

* Implemented Java EE services managing LLM API routing for email and phone-call generation, handling concurrency, rate limits, and Redis-based response caching for 1,000+ enterprise users.

* Developed a dynamic agent tooling framework where tools are defined, compiled, and attached at runtime, enabling agents to invoke APIs and backend services such as email generation, lead enrichment, and call logging, supporting ReAct style reasoning with few shot examples, enabling adoption across 10+ teams for custom agent workflows.

* Built evaluation and guardrail frameworks using Pydantic based schema validation, tracking LLM quality metrics such as hallucination rate, citation coverage, and schema failures along with business metrics like emails sent, edited, and replied, while enforcing safeguards for PII handling, spam detection, and enterprise compliance requirements.

* Drove design specs and reviews with Leads/PMs on features involving email/phone call generation flows and workflow integration, with brainstorming discussions that increased enterprise client demo adoption by 50% and accelerated production approvals.

* Designed React + Redux RTK state-management with dynamic conditional renderers, lazy-loading, pagination, and reusable UiSdl components, improving Generative Marketing UI scalability and performance by ~40%.

• Led the the development of a Gen AI chatbot for Magna International's communications team, aimed at creation of draft content for internal messages and fostering creative content generation for over 11,000 employees.

• Implemented semantic re-reranking and vector search with Azure AI Search using LangChain and deployed Hugging Face pipeline tasks along with traditional LLMs like GPT-4o, Llama-3 in Azure OpenAI for RAG process.

• Developed and deployed the application with AngularJS 17 for frontend and Python for backend utilizing Azure storage containers for data management and Azure app services and function app for deployments.

• Fine-tuned transformer models for Magna using NVIDIA tools like NeMo and Megatron-LM, employing QLora for optimized parameter storage creating a tailored LLM.

• Worked on vendor holdback application, streamlining payment processes for negative payments on damaged, excess, and overstock inventory, etc., that help Amazon reduce debit balance risk used by Finance operation analysts.

• Engineered infrastructure using AWS CDK with TypeScript, generating CloudFormation templates to streamline deployments and integrated OAuth for secure access alongside CloudWatch for real-time monitoring and alerts.

• Developed and optimized business logic using AWS SDK for Java with Apache Maven, enabling seamless integration with AWS services and ensuring scalable and reliable operations for the Vendor Holdback Service.

• Implemented microservices architecture using Java EE and Python on AWS Lambda and Fargate, enhancing dataprocessing capabilities by 60% through the use of Kafka and AWS Batch, achieving 5,000 TPS requirement.

• Architected a decoupled frontend-backend system using React and Rails APIs, orchestrating interactions with RDS- connected Java EE-based microservices, ensuring scalable and independent service development.

• Enhanced API performance for the Vendor Receivables Service by developing Spring Boot RESTful APIs, managing 10 million daily reports with JPA and Hibernate on PostgreSQL

• Mentored a software development intern and got  the opportunity to apply various leadership principles while assigning task.

• Worked on Robot teaching using Deep Learning by which robot navigates to reach Teaching Jig object.

• Robot reaches the destination in single and multiple steps using models built with CNN, LSTM, OpenCv image processing techniques

• Gained hands-on experience in applying both supervised and unsupervised learnings to tackle real-world challenges on datasets from Kaggle, UCI Machine Learning Repository and Google Dataset for comprehensive problem-solving.

• During my tenure, I honed my skills in Scikit-learn for model development, TensorFlow for implementing DL/ML operations, and Pandas for data preprocessing, Matplotlib and Seaborn for visualization and NLTK for NLP.