Avneet Bansal
Senior Delivery Consultant, AWS Professional Services · Orlando, Florida
About
I am a Senior Delivery Consultant at Amazon Web Services Professional Services, where I build cloud-native platforms and Generative AI systems for enterprise customers. My work sits where Kubernetes, infrastructure as code, and large language models meet production.
Most recently I designed an agentic contract-intelligence platform on Amazon Bedrock AgentCore. It routes each extracted field to the right model at runtime, heals itself from human corrections, and cut per-document processing cost by about 75%. The platform received the CV-S3 2026 AI-Powered Automation Excellence Award.
Before AWS, I spent over a decade in build, release, and DevOps engineering for media, entertainment, and automotive companies, and at Ygrene Energy Fund. I contribute upstream to CNCF and Kubernetes SIGs projects such as KEDA, Argo, Harbor, and external-dns.
Hint: what keeps a pod from receiving traffic it cannot serve? (9 letters)
- CKA
- CKAD
- CKS
- AWS SA Professional
- AWS DevOps Professional
- AWS GenAI Developer Professional
- Terraform Associate
- Prometheus PCA
News
- A fourth PR merged into the Amazon EKS Nitro sandbox samples: #237, a documentation note on how a burst of warm pods can land on one node, written after measuring it on a test cluster to answer a question the maintainer had opened.
- Published an adoption guide for AWS Nitro Sandboxes on Amazon EKS, with startup, kernel and egress measurements from a test cluster: One Line in a Pod Spec, a VM Underneath.
- Published textract-field-memory on PyPI and archived both libraries on Zenodo with DOIs: textract-field-memory and prompt-correction-memory.
- Published a benchmark of Amazon Textract, Bedrock Data Automation and vision LLMs on invoice photos and a 6-page invoice: Textract vs BDA vs Vision LLMs on Invoice Photos.
- A third security-hardening PR merged into the Amazon EKS Nitro sandbox samples: #215, which puts the brain pod of the brain-and-hands sample on a read-only root file system.
- Two security-hardening PRs merged into the Amazon EKS Nitro sandbox samples: #198 and #200.
- Published a migration study of Claude Sonnet 4.6 and 5.5 on Amazon Bedrock, with all numbers from fictional test data: Moving from Claude Sonnet 4.6 to 5.5 on Bedrock.
- Published a benchmark of my two memory libraries against a plain LLM on Amazon Bedrock: Spatial Memory vs Correction Memory.
- Invited lecture at NIT Jalandhar on RAG and LLM deployment in enterprise environments.
- Merged fixes into external-dns and the AWS Load Balancer Controller.
- Received the AWS Above & Beyond Award for the contract-intelligence program.
- KEDA readiness-probe fix #8226 merged.
- Won the CV-S3 2026 AI-Powered Automation Excellence Award.
- Led a webinar on scaling GenAI workloads with Amazon EKS for 237 attendees.
Open source
Libraries I authored:
- textract-field-memory, spatial memory for OCR pipelines that validates fields by position and skips LLM calls when layout confidence is high. Pure Python, zero dependencies. On PyPI, DOI 10.5281/zenodo.24150042.
- prompt-correction-memory, a correction memory that feeds human fixes back into later LLM extractions. DOI 10.5281/zenodo.24150044.
Merged upstream contributions:
- KEDA #8226, Metrics Server reports NotReady when its gRPC link to the operator is down.
- external-dns #6762, validates Route53 geoproximity bias.
- AWS Load Balancer Controller #4924, logs security group authorize/revoke failures.
- amazon.aws #3110, fixes the ec2_spot_instance IPv6 suboption.
- bedrock-agentcore-sdk-python #682, passes the boto3 session through to MemoryClient.
- EKS Nitro sandbox samples #198, #200 and #215, security hardening of the KEDA warm-pool exporter, the Docker and BuildKit build daemons and the brain-and-hands brain pod, tested on EKS Auto Mode, and #237, a documentation note on how warm pods pack onto a node, based on measurements on the same kind of cluster.
Publications
- Event-Driven Autoscaling with KEDA and Karpenter
- Human-in-the-Loop Active Learning
- Retrieval-Augmented Generation at Enterprise Scale
- From DevOps to Autonomous Platforms
- Production-Grade Kubernetes on AWS EKS
- Enterprise Cloud Transformation
- Infrastructure as Code at Scale
- LLM-Orchestrated Microservices
Talks
- Generative AI on the Cloud: RAG and LLM Deployment in Enterprise Environments
- Building and Scaling GenAI Workloads with Amazon EKS
- Orlando AWS User Group
- Amazon EKS Auto Mode workshop
Writing
- One Line in a Pod Spec, a VM Underneath: A Practical Guide to AWS Nitro Sandboxes on EKS
- Moving from Claude Sonnet 4.6 to 5.5 on Bedrock: What Broke, What Changed, and What It Cost
- Textract vs Bedrock Data Automation vs Vision LLMs on Invoice Photos: 140 Images and a 6-Page Invoice
- Spatial Memory vs Correction Memory: I Benchmarked Both Against a Plain LLM
- Your Readiness Probe Is Lying to You
- A Field Guide to Beating Amazon Bedrock Throttling
- The GitOps Bridge: Keep Terraform and ArgoCD Out of Each Other's Way
- LLM Serving on Kubernetes in 2026: What Is Solved and What Is Still Open
- Give Your Document OCR Pipeline a Memory
Awards
- CV-S3 2026 AI-Powered Automation Excellence Award
- AWS Above & Beyond Award
- AWS Customer Obsession Award
Contact
Email: avneetbansal9815@gmail.com
GitHub ·
LinkedIn ·
Scholar ·
dev.to