Headline

Software Engineer @ Visa

About

Backend Software Engineer at Visa specializing in MLOps, cloud-native architecture, and distributed systems. I currently focus on architecting and building Visa's next-generation Machine Learning Platform to AWS EKS, building a cloud-agnostic Kubeflow stack designed to serve 1,000+ data scientists. My expertise lies in developing high-performance backend services in Golang and Python, designing robust multi-tenant isolation models, and implementing complex identity federation (OAuth2/OIDC) across distributed network boundaries. Recent highlights include: • Architecting a comprehensive Kubeflow ecosystem (Notebook, TrainJob, Pipelines, KServe) on AWS EKS, implementing a custom identity federation chain for secure, zero-secret workload isolation across multiple AWS accounts. • Integrating NVIDIA DGX H100 clusters with Run:ai orchestration via custom FastAPI services to streamline GPU workloads. • Building real-time monitoring and resource allocation pipelines that reduced failed job submissions by 35% and improved backend API response times by 60%. I am passionate about tackling complex resource constraints, platform security, and building scalable, developer-first infrastructure from the ground up. Top Skills: Golang • Python • Kubernetes & AWS (EKS) • Kubeflow & MLOps • Distributed Systems • Identity & Access Management (OIDC/OAuth2)

Location

Bengaluru

AI stack

claude code

claude

codex

gemini

github copilot

mcp

Work experience

Software Engineer at Visa, Jun 2024 - Present · 2 yrs 4 mos, Bengaluru, Karnataka, India, Building and scaling Visa's Machine Learning Platform to support 1,000+ data scientists, focusing on cloud-native architecture, multi-tenant isolation, and backend orchestration. ☁️ 𝐍𝐞𝐱𝐭-𝐆𝐞𝐧 𝐌𝐋 𝐏𝐥𝐚𝐭𝐟𝐨𝐫𝐦 (𝐀𝐖𝐒 𝐄𝐊𝐒): • Architected and engineered Visa's new cloud-agnostic Kubeflow ecosystem (Notebooks, Pipelines, KServe, MLflow, Volcano) on AWS EKS from scratch, successfully replacing the legacy on-premises platform. • Implemented cross-account, per-user identity propagation between AWS SageMaker Unified Studio (SMUS) and the MLP EKS cluster. Designed a complex federation chain (AD SCIM → IAM Identity Center → EKS Pod Identity) enabling secure, user-scoped data access across VPC boundaries with zero long-lived secrets. • Designed a strict multi-tenant isolation model utilizing per-user Kubeflow Profiles, namespace-scoped RBAC, and EFS-backed persistent storage, ensuring strong workload isolation across shared GPU nodes. • Built the multi-tenant auth plane by directly integrating Microsoft Entra ID with OAuth2 Proxy, implementing JWKS validation against the EKS OIDC issuer. ⚙️ 𝐍𝐕𝐈𝐃𝐈𝐀 𝐃𝐆𝐗 𝐇𝟏𝟎𝟎 & 𝐑𝐮𝐧:𝐚𝐢 𝐎𝐫𝐜𝐡𝐞𝐬𝐭𝐫𝐚𝐭𝐢𝐨𝐧: • Developed a FastAPI orchestration service to abstract the Run:ai API ecosystem, managing complex GPU workloads on DGX H100 clusters. • Unblocked a 6-month production bottleneck by auditing and standardizing the OIDC identity bridge between Keycloak and ForgeRock. 📊 𝐏𝐥𝐚𝐭𝐟𝐨𝐫𝐦 𝐎𝐩𝐭𝐢𝐦𝐢𝐳𝐚𝐭𝐢𝐨𝐧 & 𝐃𝐚𝐬𝐡𝐛𝐨𝐚𝐫𝐝𝐢𝐧𝐠: • Architected a real-time resource monitoring system (Golang, Prometheus, Volcano Scheduler) that tracks 100+ node clusters, reducing failed job submissions by 75% and saving 20+ engineering hours weekly. • Optimized backend concurrency with goroutines and thread-safe caching, cutting response times by 60%.

Software Engineer Intern at Visa, Jan 2024 - Jun 2024 · 6 mos, Bengaluru, Karnataka, India, -> Designed and built a full-stack LLM deployment solution using React, Golang, and Python, reducing model deployment time by 90% through a seamless Docker and Kubernetes-based workflow. -> Engineered an intuitive model selection UI with RESTful APIs that supported 10+ model architectures, enabling easy integration and rapid A/B testing for researchers. -> Developed a responsive, real-time chat interface, enabling researchers to conduct interactive, qualitative evaluations of LLM conversational abilities.

Software Engineer Intern at Visa, Jun 2023 - Jul 2023 · 2 mos, Bengaluru, Karnataka, India, -> Automated the Apache Bigtop RPMs signing process, improving efficiency and reducing manual intervention per release cycle. -> Developed an automated system for comparing current RPM versions with previous versions, ensuring compatibility and identifying potential issues before deployment, increasing productivity reducing post-deployment bugs by 40%. -> Analyzed Apache Bigtop build time bottlenecks and implemented targeted optimizations, reducing build time by 20-30% for each component and saving approximately 45 minutes per build across 8+ components.

Education

Bachelor of Technology, Computer Engineering, Motilal Nehru National Institute of Technology