Mubarak (Poly4) (@poly4concept) header
Mubarak (Poly4) (@poly4concept) profile picture

DevOps Engineer || AWS Community Builder || Writes at https://teapawt.net/t.co/tucrNlsgZq

Joined June 2020
Photo by Mubarak (Poly4) (@poly4concept)Photo by Mubarak (Poly4) (@poly4concept)Photo by Mubarak (Poly4) (@poly4concept)Photo by Mubarak (Poly4) (@poly4concept)Photo by Mubarak (Poly4) (@poly4concept)Photo by Mubarak (Poly4) (@poly4concept)Photo by Mubarak (Poly4) (@poly4concept)Photo by Mubarak (Poly4) (@poly4concept)Photo by Mubarak (Poly4) (@poly4concept)
In Codelab 2 at DevFest Lagos 2026, Mubarak Alhazan (DevOps Engineer at Catering Rewards) led a hands-on session on From CI/CD to AI/CD: When Gemini Decides Whether Your Code Goes to Production. His golden rule for putting AI in the deployment pipeline: "Whenever AI makes a decision, write down what you can and let the model judge the rest." #DevFestLagos2026 #DevFestLagos
1
2
3
983
I'll be speaking at DevFest Lagos this Saturday ๐Ÿ˜Ž Come see an AI decide whether code goes to production, and help me play gatekeeper managing production Tickets at devfestlagos.com, 70% off with the code LASTCHANCE #DevFestLagos
AI is finding its way into almost every part of tech,and software deployment is no exception. ๐Ÿš€ Mubarak Alhazan, @poly4concept DevOps Engineer at Catering Rewards, will be showing us whatโ€™s possible in his #DevFestLagos session, โ€œFrom CI/CD To AI/CD: When Gemini Decides Whether Your Code Goes To Production.โ€ ๐ŸŽŸ๏ธ devfestlagos.com | ๐Ÿ“… Nov 13th & 14th
1
3
270
Last week I shared an AI deployment gate I built. It decides whether a code change ships fully, rolls out to a few users first, or waits for a human. This week I tested whether the AI was actually earning its place ๐Ÿงต
1
1
41
So I wrote that rule down as code and left the judgement to the model. With the rules in place, the models caught every dangerous change in the test. The takeaway: give the model only the judgement that can't be computed and write down the part that can.
1
3
AI is helping us write more code than ever, but most of it still ships on one signal: a green build. A README typo and a big payments change can both pass CI. Should they ship the same way if the payments service already has alarms firing? I built something for this ๐Ÿงต
๐Ÿค– Made with AI
1
1
3
197
The AI can't deploy anything itself. A separate component holds the deploy permissions. Getting it to work wasn't the most interesting part. I tested whether the AI was adding value, and what I found changed how I think about AI making production decisions.Sharing that next week
1
2
13
Most times in Infra: T๐ก๐ž ๐ญ๐ซ๐š๐Ÿ๐Ÿ๐ข๐œ ๐ญ๐ก๐š๐ญ ๐ง๐ž๐ฏ๐ž๐ซ ๐ฅ๐ž๐š๐ฏ๐ž๐ฌ ๐ฒ๐จ๐ฎ๐ซ ๐ฉ๐ซ๐ข๐ฏ๐š๐ญ๐ž ๐ง๐ž๐ญ๐ฐ๐จ๐ซ๐ค ๐ข๐ฌ ๐›๐จ๐ญ๐ก ๐ญ๐ก๐ž ๐ญ๐ซ๐š๐Ÿ๐Ÿ๐ข๐œ ๐ญ๐ก๐š๐ญ'๐ฌ ๐ก๐š๐ซ๐๐ž๐ฌ๐ญ ๐ญ๐จ ๐š๐ญ๐ญ๐š๐œ๐ค ๐š๐ง๐ ๐ญ๐ก๐ž ๐ญ๐ซ๐š๐Ÿ๐Ÿ๐ข๐œ ๐ญ๐ก๐š๐ญ'๐ฌ ๐œ๐ก๐ž๐š๐ฉ๐ž๐ฌ๐ญ ๐ญ๐จ ๐ซ๐ฎ๐ง ๐Ÿซณ๐ŸŽ™๏ธ.
4
122
Rate limiting is the first thing teams reach for against bot attacks. But attackers have adapted. distributed bots using residential proxies can spread requests across 200+ real IPs, each staying under your threshold. Your logs show nothing unusual. ๐Ÿงต
1
2
4
142
That's exactly what happened to us. Rate limiting didn't catch it, as every request looked legitimate. By the time the aggregate told the story, our email sender reputation was already taking damage.
1
2
45
The fix was using AWS WAF CAPTCHA and Cloudflare Turnstile. I wrote about how we diagnosed it, why the obvious defences failed, and how we solved it. If you're running any public endpoint that triggers an email, this is worth a read: poly4.hashnode.dev/why-rate-โ€ฆ
3
81
Just published this article on importing manual AWS resources into the Serverless Framework Stack Worth a read if you are doing IAC work: poly4.hashnode.dev/how-to-imโ€ฆ
1
3
99