Yes true. Now with all the AI hypes and these agents are just killing the human intervention in real coding .
But still any agents are still not qualified enough to understand the entire context of the product. Now it will be all about taking ownership understanding different parts of the software works altogether, understanding the entire flow. I think still no agents are good engough to handle it effectively.
It's not Twilio; it's a combination of Telnyx and Signal wire and a link trip not using their TTS & STT. I'm using a combination of Cartesia and the self-hosted misolabs for TTS. I'm using Deepgram and in some cases Sarvam as well. It's a combination of all these things.
As far as the question is why I used Telnyx and SignalWire, it's because of the cost and the latency thing. The Twilio API is not really fast as far as I've used it. I don't know about other people. I didn't find anything as such on the internet but I've used and tested both Telnyx and SignalWire. The best one is literally Telnyx but it is costly compared to SignalWire. In order to make a profit on that I had to use a combination of both.
Ok got it. Just curious, since you mentioned you are using multiple tools or combination of two or more. How are you handling the decision like which one to choose at a particular time
Depends on the country of origin of the user and also most of the tools are not, either or, they are chained together like TTS and STT is being used always with an LLM layer and signalwire when the user is in US or Canada and telnyx for other countries(Under testing)..
AI agents have to cite sources for each thing (there's a bug with displaying sources, it should let u click a fact and send you to where the agent got it. I'm working on fixing that right now). Users can also flag errors, and I'm going to periodically run fact-checking agents and manually go in and check info. However, obviously this will likely still not be perfect, accuracy will probably be the number 1 challenge with this site.
reply