Activity
Mon
Wed
Fri
Sun
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
What is this?
Less
More
3 contributions to Brendan's AI Community
Hi, I'm Ahsan. I build AI automations and I like breaking them
Hi everyone, I'm Ahsan. I'm an AI automation engineer. Before this I built Android apps for about seven years, so I care about things that have to work for a real person on a real phone. These days I build with n8n, the Claude API and Postgres. My main project is a WhatsApp booking assistant for dental clinics. It answers from approved documents and holds appointment slots. It isn't live with paying clients, so most of what I know comes from breaking it on purpose and fixing what broke. I'm open to remote work, full-time or contract. I'm here to learn how you run automations after the demo is over. What's the last thing that broke in one of your builds?
0 likes • 1d
@Mouldi Nouri Good point. In mine, old documents get marked "superseded", so search skips them. The risky case is the one you said: nobody marks the old price sheet. I want to show which chunk each answer used, so a stale one is easy to see. Do you log it in a table or in the n8n run data?
0 likes • 10h
@Mohammad Jahid Hasan Thanks, good test. In mine, a slot is held in the database first, and unused holds expire after 5 minutes. The database refuses a second hold on the same start time. Overlapping times are a different case, so two people at the same second is a test I want to run properly. Where do you keep your idempotency key, in a table with a unique rule?
Are we measuring the success of AI automation the wrong way?
I see a lot of people sharing impressive AI agents, complex n8n workflows, voice assistants, and automated systems. But here's something I've been thinking about: When you build an automation for a business, how do you actually prove it's working? Is it: - The number of hours saved every week? - The number of leads that no longer get missed? - More appointments booked? - Faster customer response times? - Or actual revenue generated? I think there's a big difference between building an automation that works technically and building one that delivers real business value. A system can run perfectly and still fail to make a meaningful difference to the business. So I'm curious: What's the one metric you use to determine whether an AI automation project is truly successful? Would love to hear from people who have deployed systems for real businesses. 👇
0 likes • 1d
Before the business number, I'd check the system number: did the thing really happen? I'd count bookings or rows at the destination, not the "success" the workflow reports. If that count is off, hours saved and leads won't mean much.
1 like • 23h
@Tony Bricks Exactly, you have rightly said it.
How do you prevent webhook replay attacks in automation workflows?
Something I've been looking into is webhook security beyond basic authentication. Even if an incoming webhook has a valid signature, what happens if someone captures that request and sends it again later? For workflows involving payments, bookings, or CRM updates, processing the same request twice could cause serious inconsistencies. A few approaches I'm exploring: - Timestamp validation to reject old requests. - Unique event IDs to detect duplicate deliveries. - Signature verification before processing the payload. - Idempotent handling so repeated events don't repeat the action. I'm curious how others structure this in n8n, especially when workflows have multiple downstream sub-workflows.
1 like • 1d
One small thing in n8n: check the signature on the raw body, not the parsed JSON. If the JSON gets rewritten even by one space, the check fails on good requests. The Webhook node has a Raw Body option for this.
1-3 of 3
Ahsan Ishaq
3
43 points to level up
@ahsan-ishaq-5187
AI automation engineer. n8n, Claude API, RAG on Postgres. Open to roles: full-time or contract.

Active 22m ago
Joined Nov 19, 2026
Powered by