Ntusekho
Mngadi

Portrait of Ntusekho looking into the camera
AI and cloud engineer in Johannesburg, South Africa AWS Certified Solutions Architect Professional

I build AI agents and cloud systems on AWS.

I’m Ntusi, a solutions architect and engineer in Johannesburg. I design AWS architectures, build the demos that prove them, and right now I’m writing AI agents in Python on Amazon Bedrock and the Claude API.

See my work GitHub profile

About

I’m Ntusekho Mngadi, Ntusi for short. I studied Computer Science at UKZN and have spent five years building on AWS: first at NTT (Dimension Data), where I went from intern to DevOps engineer to cloud architect, and now as a Pre-sales Solutions Architect at an AWS distributor. I’ve written Lambda integrations, built CI/CD pipelines, led a contact centre migration and reviewed partner architectures against the AWS Foundational Technical Review. Since passing the Solutions Architect Professional exam, most of my energy goes into getting AI agents into production.

Experience

  1. Pre-sales Solutions Architect, Cloud On DemandJan 2025 to now
  2. Cloud Architect, NTT (Dimension Data)Apr 2023 to Dec 2024
  3. Associate DevOps Engineer, NTTApr 2022 to Mar 2023
  4. Technical Accelerated Programme, NTTApr 2021 to Mar 2022
  5. BSc Computer Science, University of KwaZulu-Natal2018 to 2021

Certifications

  • AWS Solutions Architect Professional, 2026
  • AWS AI Practitioner, 2026
  • Claude 101, Anthropic Academy, 2026
  • AWS Solutions Architect Associate, 2026
  • AWS Developer Associate
  • AWS Cloud Practitioner

Also Microsoft Power BI Data Analyst Associate, Microsoft Power Platform Fundamentals, SAFe 5 Agilist and Salesforce Associate.

AI work

After passing the Solutions Architect Professional exam I went all in on getting AI into production. Most of it is Python: agent loops, tool use, retrieval, evals, and the IAM and telemetry that make it safe to run.

Built, on GitHub

AWS cost analysis agent

A tool-using agent written from first principles on the Bedrock Converse API, then rebuilt in the Strands Agents SDK. Four Python tools over live Cost Explorer data, an eval harness that took the pass rate from about 74% to 100%, and a least-privilege IAM role proven by allow and deny tests.

Python, boto3, Bedrock Converse API, Strands Agents SDK, Amazon Nova

View the repo

In progress

BrainBox

A personal knowledge assistant on the Claude API, built in phases: multi-turn chat with memory, then a RAG knowledge base on AWS, tool use and production. Along the way, structured-output tools that return JSON, like a flashcard generator and a sentiment analyser.

Python, Anthropic Claude API, Sonnet and Haiku

Architecture

Meeting intelligence tool

An event-driven pipeline that turns meeting recordings into transcripts, summaries, decisions and action items, with plain-language Q&A over past meetings using RAG.

Transcribe, Step Functions, Lambda, Bedrock, Titan Embeddings, OpenSearch Serverless

Shipped

QnABot with Amazon Bedrock

Deployed QnABot on AWS as infrastructure as code and packaged it as a partner offering, with Bedrock embeddings and LLM answers. The v7.4.0 release shipped an empty website.zip, so I traced it and hot-patched it live with v7.3.16 assets.

QnABot on AWS, Amazon Bedrock, Amazon Lex, CloudFormation

Cloud work

Five years of designing, building, securing, migrating and costing workloads on AWS.

Contact centre move to Amazon Connect

At NTT I led an on-premises contact centre migration to Amazon Connect as architect and project manager, and built the integrations around it: Lambda functions wired into Connect, DynamoDB and Lex IVR, plus custom notifications with SNS and SES. The work helped the team earn Amazon Connect Service Delivery Partner status.

Amazon Connect, Lambda, DynamoDB, Lex, SNS, SES

Pipelines and infrastructure as code

As a DevOps engineer I automated deployments with CodePipeline, CodeBuild and CodeDeploy, wrote CloudFormation orchestrated by Step Functions, and developed Python and JavaScript Lambda functions in SAFe and Scrum teams.

CodePipeline, CloudFormation, Step Functions, Lambda

Security Hub and automated response

Enabled Security Hub with the Foundational Security Best Practices standard and deployed the Automated Security Response stacks from CloudShell. Found the IAM gap blocking SSM automation, fixed it, and wrote it into the runbook.

Security Hub, AWS Config, Systems Manager, IAM

Cost dashboards across two regions

A CUDOS v5 build with billing data in af-south-1 and Glue, Athena and QuickSight in us-east-1, plus CUR 2.0 pipelines through Billing Conductor for partner cost visibility.

CUDOS, CUR 2.0, Billing Conductor, Athena, QuickSight

Migration assessments

Server discovery with the AWS Agentless Collector, EC2 right-sizing and cost scoping, then AWS Transform and Application Migration Service to move the workloads.

Agentless Collector, AWS Transform, MGN

Architecture reviews

I review production environments against the Well-Architected Framework and take partners’ software and services through the AWS Foundational Technical Review, from first review to validation.

Well-Architected, AWS FTR

Things I’m building

Better Recovery

Built by a runner with an injury, for other athletes working back from theirs. Stretches and workouts to follow, step by step, to help you recover.

iOS app, live on the App Store

Get it on the App Store

QuoteShareNow

My own site for finding and sharing quotes, served from S3 and CloudFront with small APIs behind it.

S3, CloudFront, quote APIs

quotesharenow.com

NextPage Co.

A technology and business consultancy for South African SMEs that I co-founded. I lead the AWS and solution architecture side.

Consulting, web, AWS

nextpageco.co.za

SA telecoms on Twitter

UKZN group project that classified South African telecoms’ Twitter footprint before and during Covid-19.

Neo4j, Java

View the repo

Off the clock

Running

Half marathons, and Car Chats: a minute talking to camera, then the run, posted to TikTok.

Handstands

Chasing a freestanding handstand. Hips over head is the goal. The banana back is still winning.

People

Going out, meeting new people and figuring out why people do what they do. Human psychology fascinates me, and so does a night of laughing with friends.

Guitar

Learning neo-soul fingerpicking with Tom Misch, Steve Lacy and Daniel Caesar on repeat.

Aeroplanes

Aeroplanes and aerospace pull at me. One day I want to work closer to that industry.

Science

How things work, and going down rabbit holes until they make sense.

Ntusekho laughing in a striped shirt and jeans