Applied AI & SRE Manager · Soar Inc.
Updated Nov 2026

I run engineering with AI agents on the team, and a human at the production gate.

I manage applied AI and site reliability at Soar Inc., a SAMA-regulated fintech: five pre-production environments, regulated production, and a delivery pipeline of AI agents that I designed and gate. Cairo ↔ Riyadh.

How I run engineering

Agents do the work. You hold the gate.

This is the path one change takes. It stops when it reaches you.

one change, request to production Illustration
  1. 01 Request
  2. 02 Plan
  3. 03 Build
  4. ↩ sent back once 04 Review
  5. 05 Approve
  6. 06 Verify
A colleague asks for a change.

Every request becomes a tracked card: 155 from 22 colleagues in five weeks.

Decisions

What I decided, and what happened.

01

Rebuild instead of patching

Case study →
Context
An audit of 88 services showed 73% would silently undo a fix on their next deploy.
Decision
New isolated environments and one release pipeline.
Result
15 repositories on one release path. 29 legacy workflows removed.
02

Delete my own platform

Case study →
Context
About 390K lines of in-house agent platform. One tool made 93% of its changes.
Decision
Replace it with a managed agent and repository rules.
Result
Custom gateway retired after a zero-traffic window, with a one-command rollback kept.
03

Refuse it twice

Case study →
Context
An internal monitoring product looked finished.
Decision
Rejected two versions, then ordered two independent audits of the live build.
Result
30 of 78 displayed fields were wrong. Fixed before sign-off.
A test that cannot fail is worse than no test.
One of six rules the team works by
Experience

The road here.

Soar Inc. · RiyadhTwo roles since 2024
2025–now

Applied AI & SRE Manager

  • Lead a team of four (since 2025) in applied AI and site reliability for a SAMA-regulated fintech: 5 pre-production environments, 130 services, and regulated production.
  • Designed and run an AI-agent delivery pipeline: 744 pull requests merged across 44 repositories in ten weeks.
  • Built a governed AI DevOps agent: 338 changes in eight weeks, each approved by a person or a written policy.
  • Moved product environments off shared clusters: 15 repositories on one release pipeline, 29 legacy workflows removed.
  • Cut cancelled deployments from 79 in about 44 hours to zero.
  • Wrote the hiring kit for two senior roles and a 24-page handbook for new engineers.
  • Shipped the company's first production agents: a QA agent that filed its own bug reports and a bug-fix agent that opened pull requests.
2024–2025

Cloud / DevOps Tech Lead

Led a 6-person team through a SAMA audit to 100% compliance, and cut time to recover from 2.5 hours to 20 minutes.

Earlier roles
2024

Platform / DevOps Consultant Enterprise client · via Soar

Delivered a workforce capacity-planning platform for an enterprise client and ran its CI/CD, infrastructure as code and observability.

2023

AWS Senior DevOps Engineer PwC

Owned CI/CD and multi-cloud infrastructure for a PwC UK platform.

2021–2023

DevOps & Full-Stack Engineer JATDEV

Ran CI for a microservices platform and built a WebRTC media server for multi-party calls.

2021

Enterprise Systems Developer Mideast Software

Built ERP modules and REST APIs, deployed with Docker Swarm.

2018–2021

Software Developer Enterprise Consultancy Services

Developed and maintained point-of-sale systems for retail clients.

2017–2018

Java Backend Developer (Junior) 3M Defense

Built and maintained Java backend services and REST APIs.

Merge, then prove it is working.
Another of the six
Leading people

The team around the agents.

Hiring
Wrote the role profile and a 45-minute scenario interview, built from the team’s own incidents, for two senior roles.
Onboarding
Published a 24-page estate handbook and onboarded a senior engineer across 8 platforms in one morning.
Enablement
Rolled out per-engineer cloud workstations for running AI agents.
Stakeholders
Wrote the production go-live readiness assessment in two registers: technical, and plain language for executives.
Skills

Leadership first, tools last.

Leadership and delivery
Hiring and onboarding · Stakeholder briefings · Incident review · Delivery pipelines · Cost attribution
Applied AI
Agent orchestration · Human-approval design · MCP · Multi-model orchestration · Mutation testing
Reliability and observability
SLOs and alerting · OpenTelemetry · Grafana · Splunk · Incident response
Platform and cloud
Google Cloud (Cloud Run, GKE) · AWS · Azure · Alibaba Cloud · Terraform · GitHub Actions · Managed Kafka · WebRTC and SIP
Security and compliance
SAMA-regulated delivery · Keyless CI authentication · OAuth 2.0 and OIDC · Audit trails
Languages and data
Python · JavaScript / Node.js · Go · Bash · MongoDB · PostgreSQL · Redis
Projects

Things I shipped.

Soar Build

An in-house work-management platform that replaced Jira for the company.

Next.js · Hono · Postgres

Perform Plus, PwC UK

CI/CD and cloud infrastructure for a hybrid-work platform.

Azure · GCP · Terraform

SpaceJat

An event-driven collaboration suite with a WebRTC media server.

Nuxt · Socket.io · Redis
Receipts

Every number, with its source.

Tap a line to see what was counted, where it came from, and my part in it.

Counted
Pull requests opened by the agent pipeline and merged.
Source
GitHub, searched by merge date.
My part
I designed the pipeline, direct it and hold the merge gate. I did not hand-write these.
Counted
Change pull requests opened by the DevOps agent across 21 services and 5 environments. Median 4 minutes from proposal to merge.
Source
GitHub, the service-definitions repository.
My part
I built the agent, the approval rules and the audit trail. None of these touched production.
Counted
Running Cloud Run services, plus 25 worker pools.
Source
Cloud inventory, read the same day.
My part
I led the move from shared clusters to isolated environments on one release pipeline.
Counted
Review rounds that ended in changes requested, across 576 work items.
Source
The delivery board's run log.
My part
I set the rules: the reviewer is never the author, and every guard must be shown to fail.
Counted
Cards on the delivery board, one per piece of work.
Source
The board's own records.
My part
I set up the board, the roles and the hand-offs. Agents work the cards.
Counted
Requests that arrived by chat or ticket and were filed as cards.
Source
The delivery board's records.
My part
I designed the intake and the triage. This counts requests captured, not requests closed.