I’d be willing to review your voice agent. I’ve been building contact center solution with various other applications for over 25 years. I’ve been focusing on getting really well versed in Retell AI.
Not sure if you are using the conversational flow builder or single prompt. I’m using my past experience with similar type conversational flow workflow tools to build the best solutions.
I have built and optimized several voice agents in Retell AI, covering everything from prompt structuring and node transitions to voicemail detection and IVR handling, so the areas you mentioned are very familiar to me.
Refining conversation flow and reducing unnatural pacing or filler is usually where the biggest quality gains come from, and it sounds like that is exactly what you are after.
I can take a look at your current setup and give you practical, specific feedback rather than just general advice.
Hey @kkean, we’d love to connect. I run Agento AI, a certified Diamond-tier Retell partner, and optimization work like this is most of what we do.
Your list maps almost exactly to problems we’ve worked through on live client agents: converting flows to live transfer and fixing the failure paths, replacing IVR-style menu logic with natural context inference, and prompt architecture that stays maintainable as node counts grow.
Happy to hop on a Zoom, walk your flows, and come back with specifics!
@kkean — you already have several strong implementation offers here, so I’ll come from a different angle.
MARUPA focuses on independent Voice AI QA rather than implementation.
For an existing agent like yours, I’d establish a baseline across the current workflow, then test realistic edge cases and failure paths before and after any optimization work.
The goal is to verify whether the agent actually improved — not just whether the prompt or flow looks cleaner.
That can include conversation quality, transfer behavior, failure handling, data preservation, and whether the intended business outcome actually happened.
I can start with a small QA Snapshot on the existing agent so you can evaluate the findings before deciding whether it’s useful.
@kkean this thread has been open four months. That usually means the offers were opinions, & an opinion cannot be checked.
Different approach. I write your edge cases as scenarios & run them against the live agent. Voicemail detection, IVR navigation, failed transfer. Each one comes back pass or fail. Not a note.
I run a production voice stack. Warm time to first audio sits near 1.5 seconds. The unqualified-lead gate passes 4 of 4 & the abuse hard-stops pass 3 of 3, checked on every deploy.
Conversation naturalness is where I was wrong the longest. I tuned tone by ear & could not tell you whether a change helped or hurt. That is why the harness exists. The argument ends with a number now instead of a preference.
The reliability question is the same shape. A transfer that worked Tuesday breaks Thursday & nothing about the transfer changed. You find out because a caller tells you, which is the worst way to find out.
You keep the harness. Change a prompt, re-run it, know in a minute whether you broke a transfer.
Want me to run five of your real scenarios first so you can see the format? DM me here.
@kkean If this is still open, I would review the live agent against the production paths you named, then return a reproducible failure list instead of a prompt opinion.
Relevant proof: my current voice path reaches about 1.5 seconds warm time to first audio. The platform behind it has 76 edge functions and about 140 Postgres tables. Its release gates currently pass 4 of 4 unqualified lead cases, 3 of 3 abuse stops and 12 of 12 AI CRO eval suite.
I can review call flow, prompt behavior, transfers, tool calls, fallback behavior and edge-case coverage. I will not quote scope or timing until I see the current workflow and the failures you want fixed.
Message me read-only access and I will tell you whether I am actually a fit.
Hi @kkean — Damian here, founder of Natyv AI (US Eastern). I can audit and tune an existing Retell agent end-to-end: call flows, prompts/nodes, edge cases, qualification and booking, plus SMS follow-up. Happy to review the current setup and scope a focused paid engagement. Contact Damian@get-myagent.com or https://natyv.ai.
Kevin, voicemail, IVR navigation and failed transfers are good acceptance tests because each can be returned as evidence, not a prompt opinion. I propose a fixed $450 diagnostic: five agreed scenarios, PASS/FAIL/BLOCKED results, timestamped failure evidence, and the three highest-impact reliability fixes within one business day. No production changes in the diagnostic. If the need is still live, can you fund the milestone and provide a callable test agent or read-only access? Email kalin.smolichki@gmail.com or DM here.
Saw your post about refining your live Retell agent, particularly voicemail detection, IVR navigation, failed transfers, and conversation quality. We can help review the workflow, identify failure points, and test fixes against real call scenarios.
I run Pinnakle Media. We build AI voice agents on Retell and Vapi and connect them to the tools you already use, like GoHighLevel, n8n and your calendar.
Here’s how I’d handle yours:
Go through your call flow and find where calls break
Build or fix the agent so every call gets booked, transferred or logged the right way
Test it with real calls before it goes live
If it helps, we can start with a small paid test, so you see the work before you commit to the full build.
Open to a 15-minute call this week? You can grab a time here: