Hey! I'm Harshil Yogi, an AWS DevOps and Automation Engineer.

Engineers had to manually dig through CloudTrail, VPC Flow Logs, ALB logs, and custom application logs scattered across multiple S3 buckets to investigate incidents.

Logs sit in AWS's native delivery layout across multiple S3 buckets in JSON, CSV, JSONL, and gzip-compressed formats, so the tool had to auto-discover and parse them with zero pre-processing or data migration.

My role: End-to-end architecture and delivery: the Lambda, API Gateway, and Bedrock pipeline, fully deployed via Terraform.

A Lambda function behind API Gateway auto-discovers and parses CloudTrail, VPC Flow Logs, ALB logs, and custom application logs directly from S3, then hands the parsed context to Bedrock (Claude) to answer the engineer's question in plain English.

Manual EC2 load monitoring and manual AWS Inspector v2 vulnerability reporting were consuming engineering time.

Two workflows had to run end to end without manual steps: pattern-based EC2 CPU alerting (70%/95% thresholds with 4-spike alerting logic) and recurring vulnerability reporting.

My role: Architecture design, Terraform provisioning for the full stack (separate state files per use case), Bedrock integration, and dashboard setup.

EventBridge polls EC2 CPU metrics every 2 minutes; four Lambda functions detect sustained-load patterns and hand them to Bedrock Claude 3 Haiku, which writes state to DynamoDB and triggers SNS alerts. A second, monthly-scheduled path pulls AWS Inspector v2 findings into a serverless API Gateway dashboard. Both paths deploy via Terraform with separate state files.

The client's production Windows Auto Scaling Group ran outdated AMIs. Every patch cycle was fully manual. There was no automated way to patch, rebuild the AMI, and update the Launch Template without touching existing config.

My role: End-to-end design and delivery: the CloudFormation template, the Image Builder pipeline, and the Lambda automation.

A single CloudFormation template combines EC2 Image Builder with SSM Patch Manager. It patches the instance and builds a new AMI. A Lambda function then updates the ASG Launch Template automatically. EventBridge triggers the whole pipeline weekly, Sundays at 2 AM UTC.

The client needed a full affiliate cashback platform with automated payouts and wallet management, without manual operations overhead.

Payouts had to stay accurate and traceable while removing manual operations from the loop.

My role: Automation layer and backend integrations.

n8n workflows drive cashback tracking, claim approvals, wallet updates, payout execution, and notifications. Supabase holds wallet and state data behind a Next.js app.

Manual lead management was slowing response time and hurting conversion.

Every lead needed deduplication, scoring, and follow-up without a human touching the pipeline.

My role: Full automation design and build.

n8n captures leads from website forms, deduplicates them, and scores them with Google Gemini. It sends personalized follow-ups, instantly notifies the owner for high-value leads, generates daily reports, and creates a CRM deal automatically on conversion.

Content creation from engineering activity was entirely manual. LinkedIn posts had to be written by hand.

Auto-generated drafts could expose sensitive engineering details, so publishing needed a human approval gate.

My role: Infrastructure (ECS Fargate, Terraform) and approval workflow design.

The system converts GitHub commits and Slack discussions into draft LinkedIn posts. A Slack approval step gates every post before it publishes. It runs on ECS Fargate, deployed via Terraform.