Vinoth Kumar Ganesan

• Architect and develop enterprise-scale GenAI solutions, including Agentic AI and Multimodal RAG
platforms, to support merchant operations and decision intelligence.
• Design and implement LLM fine-tuning and distributed training workflow using NVIDIA NeMo,
Megatron Core, and PEFT/LoRA techniques.
• Build and optimize retrieval, inference, and evaluation pipelines leveraging LangSmith, RAGAS,
and LLM-as-a-Judge for continuous performance monitoring.
• Architected a multi-agent AI platform for autonomous customer support and merchant issue
resolution using Redis checkpoint memory/caching, semantic caching, and Agent-to-Agent (A2A)
communication via Model Context Protocol (MCP).
• Implement AI safety, governance, and Human-in-the-Loop (HITL) frameworks using NeMo
Guardrails and controlled deployment workflow.
• Lead end-to-end platform development, collaborating with cross-functional teams to deliver
scalable, production-ready AI solutions for payments and merchant ecosystems.
Key Achievements
• Reduced production retrieval latency by 50% by fine-tuning Llama 3.1 Nemotron with
LoRA/PEFT and implementing HyDE-based embedding generation.
• Delivered $7M+ annual cost savings by architecting an AI agent-driven interchange
optimization platform that proactively detects and remediates payment degradations.
• Automated 600+ hours annually by building a Multimodal RAG system with A/B testing,
advanced reranking, and HITL approval workflow.
• Reduced customer support resolution time by 40% and query latency by 35% by deploying
a multi-agent Agentic AI platform with Redis semantic caching, DeepEval pipelines, and NeMo
Guardrails-enforced compliance.
• Led development of an AI-powered sales intelligence platform that improved merchant analysis
efficiency and enabled churn prediction and personalized growth strategies.

• Designed and deployed machine learning models for risk analytics, fraud detection, and credit
decisioning within banking environments.
• Developed end-to-end ML pipelines using TensorFlow and Scikit-Learn, including feature
engineering, model training, validation, and deployment.
• Owned fraud detection model lifecycle, monitoring performance and continuously improving
detection accuracy across transaction channels.
• Implemented MLOps practices, enabling scalable deployment and operational stability through
containerization and CI/CD automation.
Key Achievements
• Reduced loan by 35% approval model
• Prevented $2M+ in fraud losses by delivering a fraud detection system with 94% transactions.

• Developed RFDS platform (Angular 4, Java 8, Spring Cloud) supporting 500+ cell sites.
• Improved database performance by 60% through RESTful microservices with Hibernate.
• Implemented CI pipeline, reducing release cycle from 2 weeks to 3 days.

• Built Contrax contract management system (Java/J2EE, AngularJS) managing 1, 000+ contracts
and $50M+ in service value.
• Integrated SharePoint API, reducing document retrieval time from 5 minutes to 15 seconds.
• Enhanced database performance by 3x using Spring and Hibernate optimization.

• Developed enterprise Java applications and back-end services supporting large-scale business
operations.