Ajas Bakran

I’ve been diving into tools like Voiceflow and Retell, testing real call flows and latency.But most tutorials I find are focused on sales strategy and mindset, not the technical side — the integrations, logic, APIs, and data flow that make a voice agent actually work in production. 👉 Do you know any solid learning resources that go deep into the technicalities of building voice agents — not just the sales process? Would love to build a proper roadmap for mastering this craft.

New comment 24d ago

Ajas Bakran

0 likes • 27d

The API docs themselves are usually the most helpful starting point. For Retell and VAPI specifically, their developer documentation walks through webhook structures and function calling patterns which is really where the magic happens in production. Once you understand how those platforms pass context back and forth, you can map out data flows and error handling before writing any code. Testing latency in different network conditions is another thing worth doing early, especially for call routing scenarios.

Dylan Yeo

28d •

Support Needed 💻

Which Voice AI platform for my use case?

Hi guys, I'm new to building Voice AI agents so I would need help. I'm currently building a custom project for my client. I've attached a diagram so that it's easier to visualise. Let's keep it simple for now and just focus on the Voice AI section: He essentially wants a simple high level agent that decides the intent of the inbound call. Customer looking for support --> Customer Support Agent Customer interested in their services/product --> Sales Agent Customer has billing issues --> Financial/Admin Agent ... and maybe more agents (potential for scale) I think my main decider between all the Voice AI platforms available would be the agent transfer/routing feature, ease of use, and maintainability. Here are my thoughts and experience so far: Vapi - Overwhelming. Really developer centric. I think I have to manually integrate RAG on n8n/make.com? Can't seem to make routing and transfering work for some reason. ElevenLabs - So far the new Workflow feature does what I want. Just slightly clunky at the moment. Might be too pricey for my client. Retell AI - Has agent transfer feature. Probably by far the most promising of them all. Synthflow - Not sure how to get multi-agent thing to work just yet, or you probably can't? But you can use one agent for Voice/WhatsApp/Email/Widget. Curious what you guys think. Do let me know if there's anything else I need to clarify :)

New comment 26d ago

Which Voice AI platform for my use case?

Ajas Bakran

2 likes • 27d

Your instinct on Retell is solid for intent based routing. One thing to test is how your client wants handoffs to feel. If they need truly seamless transfers without the caller noticing, check how each platform handles context passing between agents. Some platforms reset conversation history on transfer which breaks continuity. Also consider whether your client needs real time call monitoring and analytics per agent type, that can help them optimize which calls go where over time. Might save them from routing errors early on.

Minh Khang Dinh

Oct 28 •

Support Needed 💻

Need help with App Dev

Hey guys, I need help with an app we are building. We are implementing RAG to an app. Also with functions like uploading files, etc. Client has problems with uploading and testing the RAG in the app but I think the problem might be in supabase permissions or sth. I need someone experienced with app dev to look into this.

New comment Nov 1

Ajas Bakran

0 likes • Nov 1

Building on what Hicham and Aanchal mentioned - if uploads succeed but RAG can't read them, check if your pgvector extension is enabled in Supabase and your embeddings table has proper RLS rules. A common gotcha is that the storage bucket allows uploads but the embeddings retrieval function runs with different auth context and can't query the vectors table. Make sure your service role key is being used for the RAG pipeline, not the anon key.

Jose Pastrana

Oct 27 •

Support Needed 💻

Error: Pretty-print {"status":"error","message":"Unauthorized"}

Hi AIS fellow, Is anyone else getting attached error in n8n? This happens when adding any credential in N8N. Pretty-print {"status":"error","message":"Unauthorized"} I had this error in the past, a quick log-off and/or restart of N8N instance fixed it. But now it simply won't go away.

New comment Nov 1

Error: Pretty-print {"status":"error","message":"Unauthorized"}

Ajas Bakran

0 likes • Nov 1

If clearing cache and checking OAuth redirect URLs doesn't solve it, the issue might be deeper - if you ever changed your N8N_ENCRYPTION_KEY environment variable or restored a database from a different n8n instance, the stored credentials can't decrypt properly. n8n can't tell you this directly, it just throws Unauthorized. Worth checking if your encryption key is consistent, especially if you recently migrated servers or restored from backup.

Muhsin B

Oct 29 •

Support Needed 💻

Instagram Comment to DM Automation

Does anyone know how to configure the HTTP Request node to send the DM after you get a comment with a keyword. I understand there's manyChat etc, but cant we do it through n8n or make.com?

New comment Nov 1

Ajas Bakran

2 likes • Nov 1

The Instagram Graph API requires your app to use the instagram_manage_messages permission, and even then, you can only send messages in response to someone who messaged you first (user-initiated conversations). Sending DMs from comments alone isn't allowed by Meta's API, which is why ManyChat and GHL are recommended - they're official Meta Business Partners with special API access. If you try to bypass this with unofficial methods, your account will likely get flagged. Callum's right about sticking with the official tools for this specific use case.

1-10 of 31

Level 3 - AI Operator 🛠️

10points to level up

Ajas Bakran

@ajas-bakran-4065

Expert Voice AI Agent Developer | Helping Business to Unlock the growth

Active 4h ago

Joined Mar 20, 2025