Amit Yadav

Focus: Agentic AI Systems & GCP AI MlOps Engineering

• ● Collaborated with Google's Core Generative AI team on Agent Development Kit (ADK) deployment frameworks.

• ● Designed and deployed multi-agent systems for cybersecurity clients using ADK and A2A protocol and MCP

• ● Integrated LLM observability tools (Arize, W&B, MLflow, AgentOps) for production-level evaluation.

• ● Deployed agent systems on Vertex AI Agent Engine and Cloud Run.

• ● Architected RAG pipelines using Vertex AI RAG Engine, GCS, and memory bank.

• ● Built scalable LLMOps pipelines within GCP using GKE and GitHub Actions.

• ● Provisioned GKE infrastructure and implemented IAM, IAP, and service account security models.

• ● Delivered multimodal agentic systems integrating Tavus and HeyGen with ADK Live API.

• ● Designed end-to-end MLOps platform with:

◦ o Kubeflow(Integrated with Vertex AI pipelines)

◦ o Load balancers

◦ o CI/CD automation

◦ o Model monitoring & drift detection

◦ o Staging-to-production deployment pipelines

• ● Leading the team, and working as individual contributor both

• ● Designed cross-cloud MLOps platform, deployed across GCP, Azure, and AWS, Deployed Custom kubeflow Dashboard with all its components, added mlflow as other component to ensure proper AI/ML model loggings and tracking, Deployed ArgoCD etc opensource packages.

• ● Orchestrated Kubeflow pipelines for scalable training and deployment.

• ● Built GitHub Actions CI/CD workflows for ML lifecycle automation along with github runner.

• ● Provisioned infrastructure using Terraform across multi-cloud and deployed Kubernetes clusters on GKE, EKS, and AKS.

• ● Implemented secure API gateways (APISIX), Istio for ML microservices and integrated enterprise monitoring stack (Prometheus, Grafana, ELK). also Containerized ML workloads using GCR, ACR, and ECR.

• ● Designed and implemented an end-to-end ML pipeline using KFP v1 on Vertex AI Pipelines, covering data extraction, validation, preprocessing, training, evaluation, and deployment parameterized to serve both classification and regression use cases from a single codebase

• ● Built custom containerized training jobs (AI Models, Time series models) on Vertex AI Training with Experiment tracking, and implemented a shared preprocessing layer ensuring consistent feature transformations across train, eval, and serving stages

• ● Implemented a champion vs. challenger model governance pattern with configurable promotion thresholds, versioning in Vertex AI Model Registry, and metric labels for programmatic champion identification

• ● Developed a FastAPI serving container deployed to Vertex AI Endpoints with auto-scaling, and configured Vertex AI Batch Prediction jobs with BigQuery as source and sink

• ● Configured Vertex AI Model Monitoring v2 for training-serving skew and prediction drift detection, with per-feature thresholds, prediction sampling, and email/Pub-Sub alerting

• ● Built a two-stage CI/CD system using GitHub Actions (lint, tests, pipeline compile gate on PRs) and Cloud Build (Docker build → GCS artifact push → Vertex AI pipeline submit on merge), with keyless GCP auth via Workload Identity Federation