We are initiating a structured implementation phase to evaluate execution quality, system reliability, and delivery standards ahead of a potential long-term collaboration with **Levanrohe.com**. This phase requires building a **complete, production-grade reservation automation pipeline**, starting from AI voice interactions and ending with operational notifications to restaurant staff. The objective is to validate that the system can be delivered **cleanly, deterministically, and without manual intervention**. Below is the required scope and execution logic. --- 1. Objective (Non-Negotiable) Build an end-to-end automated flow that: 1. Receives finalized reservation data from **Retell AI** 2. Processes and validates that data in real time 3. Pushes the reservation into **xMenu** in a structured, API-compliant manner 4. Notifies restaurant staff via **Zadarma SMS** 5. Operates entirely through **Make.com** as the orchestration layer This system must function reliably under real operational conditions. --- 2. Authoritative Flow Overview **Source of Truth:** Retell AI (call-ended event) **Orchestration Layer:** Make.com **Downstream Systems:** xMenu (data sink), Zadarma (notifications) High-level flow: ``` Customer Call → Retell AI (call concludes) → Make.com Webhook (authoritative payload) → Data validation & normalization → xMenu API (reservation / order write) → Zadarma SMS (staff notification) ``` No UI-driven steps. No manual confirmations. No hidden dependencies. --- 3. Step-by-Step Technical Responsibilities 3.1 Retell → Make.com (Webhook Ingestion) * Configure Retell to send **call.ended** events to a Make.com webhook * Payload must include: * Guest name * Phone number * Date * Time * Party size * Notes (optional) * Treat this payload as **finalized reservation intent** Make.com must reject incomplete or malformed payloads explicitly. --- 3.2 Data Validation & Normalization (Make.com) Inside Make.com: * Validate required fields

One thing I’ve noticed while working on AI visuals is that not all prompts are equal, even if they describe the same idea Most people use AI image tools with a normal text prompt like: “Create a premium, editorial-style image for a brand.” That can work — but results are often inconsistent Here’s why - A normal prompt is more of a suggestion You describe what you want and the AI makes a lot of decisions on your behalf A structured (JSON-style) prompt is different It gives the AI clear rules and priorities Think of it like this: A normal prompt says: “Make this look luxury.” A structured prompt defines: • lighting • camera angle • materials • environment • what must stay consistent • what must not change Same idea — very different output quality Do you need to write structured prompts yourself? Not really In practice, structured prompts are best generated by an AI chatbot, not written manually They’re easy to break, and one small mistake can cause parts of the prompt to be ignored. The smoother workflow is: 1. Explain your intent in plain language 2. Let an AI convert that into a structured prompt 3. Use that prompt in the image model This approach leads to: • more realistic images • better consistency across campaigns • fewer “AI-looking” artefacts • more control for brand work What’s worked best for me From testing different tools: • Gemini / Nano Banana handles structured prompts particularly well • Normal prompts are great for exploration and ideas • Structured prompts are better when you need repeatable, professional visuals Especially for websites, ads or launch campaigns, structure makes a big difference Most users never need to think about this layer, but once you understand it the quality jump makes sense Have you ever used JSON?

We have chosen only a handful of the best applicants, to be buildmyagent testers. (If you applied, check your email. If you don't have it, check spam as well) Here's the deal: - For every 5 bugs you find (you get a free month of buildmyagent) If you find 1 bug, you also get a free month. This can be small UI bugs, or larger logic bugs (Check your email for full instructions) And remember to join the tester discord as well Let's make this the best software possible 🚀