Client Work

Six client projects.
What each one delivered.

Representative projects across media, insurance, pharma, hospitality, and financial services, each one built by me. Client names withheld per NDA.

Book a 15-minute call with Charles See the services

A row of steel gantry cranes along a container quay at blue hour, one lit with amber work lights, still water in front.

Client Engagements

Industries Served

Repeatable Playbooks

Years Cloud Experience

The Work

Each project is shown as the client saw it: the problem, what I built, and what changed.

An empty film soundstage with a grid of steel lighting truss overhead and a single amber spotlight beaming down.

Media & Entertainment

Content moderation pipeline and recommendation engine

Data & AI Engineering

Challenge

A client building a mobile media platform needed a content moderation pipeline and a custom recommendation engine that learned from user interactions.

Solution

I built a Kinesis-to-S3 signal pipeline with Glue ETL for interaction data
I orchestrated content moderation with Step Functions (Rekognition, Transcribe)
I trained a SageMaker recommendation model on an OpenSearch vector database
I provisioned all of it with AWS CDK

Processing time fell from 4–6 minutes to under 30 seconds once the moderation steps ran in parallel.

A law library aisle at night, tall shelves of plain bound volumes and a single amber reading lamp on a long table.

Insurance

Automated legal bill review

Data & AI Engineering Automation & DevOps

Challenge

A legal team reviewed 5% of incoming law firm bills by hand for contract compliance, so overbilling on the other 95% went unchecked.

Solution

I extracted contract rules automatically with Claude 3.5 Sonnet on Bedrock
I read the PDF invoices with Textract
I verified each bill with Lambda and Step Functions
I built a Flask web UI for human review, all provisioned with Terraform

Every incoming legal bill is now reviewed automatically in real time: 100%, up from 5% by hand.

Pharma

Fleet-wide compliance inventory and remediation

Governance & Compliance Automation & DevOps

Challenge

A pharmaceutical client needed a daily inventory of every AWS resource across 100+ accounts to verify compliance, with automated remediation.

Solution

I wrote a Python application with a plugin architecture, run as a Step Functions state machine
It scans every account and region in parallel
It remediates automatically: VPC flow logs, Route 53 query logging, S3 encryption
It also enforces backups and tag compliance and automates the reporting

The inventory runs nightly across 100+ accounts in under one hour. The client created a dedicated internal team to manage and extend it.

A climate-controlled archive vault with rows of dark steel shelving receding toward a single amber lamp.

Pharma

Two-petabyte business continuity

Cloud Infrastructure & Architecture Governance & Compliance

Challenge

A pharmaceutical client urgently needed a business continuity copy, isolated from production, of 2 petabytes across 600+ S3 buckets in 100+ accounts and 14 regions.

Solution

I scripted the inventory, versioning, and encryption in Python
I set up cross-account replication to isolated destination buckets with Object Lock
I added Glacier Deep Archive lifecycle policies for the long-term copy
I orchestrated batch replication across every account and region

All 2 PB reached isolated destination buckets under Object Lock in about 50 hours.

An empty restaurant kitchen at night, a long stainless-steel pass under a single amber heat lamp.

Restaurant / Hospitality

Migration Acceleration Program assessment

Cloud Infrastructure & Architecture

Challenge

AWS asked for a Migration Acceleration Program assessment of a point-of-sale platform running on EKS, Aurora, SQS, and on-premises Kubernetes.

Solution

I wrote a six-phase migration plan, with Control Tower governance first
I removed the WebSocket bottleneck and secured the API with API Gateway
I planned the move from Aurora to DynamoDB global tables
I designed multi-region high availability with CloudFront and Route 53 latency routing

The client received a 6-phase plan across 2 regions, with an expected monthly cost below its current AWS bill.

A financial district skyline at blue hour reflected in still water, one tower's crown lit amber.

Financial Services

Real-time market prediction pipeline

Data & AI Engineering

Challenge

Quantitative analysts needed real-time ingestion of financial data, plus model training and inference for market predictions.

Solution

I streamed the Bloomberg API through Kinesis with WebSocket ingestion
I built online and offline feature stores for the training data
I trained on A100 GPUs in SageMaker and served real-time inference over socket.io
I added sentiment analysis from news feeds and press releases

Continual online training time is down to about 60 seconds, with live prediction streams.

Sample case studies

Four engagements, start to finish.

Real clients’ findings and bills stay confidential, so I ran four of the six-week engagements for three fictional clients, StrataForge Robotics, Seatlark, and Copperwick Therapeutics, and published everything: every week, the workspace that held it, and every report.

A row of low autonomous mobile robots parked at their charging docks down a long warehouse aisle at blue hour, amber lamps glowing on the nearest machines.

Sample · fictional client

AWS Cost Analysis & Optimization

StrataForge Robotics: the Optimization, start to finish

A fictional robot maker with a $55.7 million AWS bill: $23.5 million a year identified, $11.1 million put in place, and $9.4 million measured on its own Cost and Usage Report. Every week, the Pavilion workspace, and all eight reports.

Read the sample case study

An empty stadium concourse at blue hour before the gates open: a long row of steel turnstiles receding into haze, one with a small amber indicator lit.

Sample · fictional client

AWS Cost Analysis & Optimization

Seatlark: the Optimization, in the middle of a migration

A fictional ticket exchange at about $50,000 a month, eleven months into a move out of a colo data center: $246,839 a year identified, $148,967 measured on its own Cost and Usage Report, and the gate cleared 4.35 times. The bottom of the engagement’s range, where the gate is hardest to clear.

Read the sample case study

Three brushed-steel liquid-nitrogen storage dewars in a dark cold room, white fog rising from their lids in steel-blue light, small amber status lamps lit on two of them.

Sample · fictional client

AWS Cost Analysis & Optimization

Copperwick: the Optimization, under GxP change control

A fictional mRNA medicines company with validated systems: $4,310,044 a year identified, net of its private pricing discount, $2,401,346 measured on its own Cost and Usage Report, and eight validated-system changes that wait on change control and a freeze instead of skipping them.

Read the sample case study

A long aisle of steel-mesh parts cages in a robotics plant at blue hour, one cage door standing open under an amber lamp.

Sample · fictional client

AWS Security, Compliance & Governance Audit

StrataForge Robotics: the Hardening, start to finish

A fictional robot maker with 48 AWS accounts: 54 findings verified, nine of the ten criticals closed in two days and the tenth accepted in writing, and the same scan from 32.4% to 91.1%. Every week, the Pavilion workspace, and every report.

Read the sample case study