I switched to Haiku and like it but when it gets to numbers it’s a mess. It repeats zip code like “fifteen two twenty” instead of one five two two zero. Same with phone numbers. I didn’t have any issues with GPT4.1 but wondering if there’s a fix for this?

Apologies for all the questions, I'm still finding my feet and I believe there is seasoned veterans here whose brains I have the chance to pick. Based in Ireland, building a receptionist agent for a clinic. Trying to get her sounding natural and colloquial rather than robotic. What have you guys found works best - hardcoding specific filler words and phrases, leaving it open, or does trying to manufacture it actually make things worse? Any other general tips to get the most out of voice agents would be appreciated

Working on a build for a healthcare client. When a patient calls in, the agent needs to know who they are and pull basic details (name, upcoming appointments, etc) Two approaches I’m weighing up: 1. Bulk import all patient contacts into GHL ahead of time so the agent can match on caller ID instantly 2. Real-time tool call / webhook that hits the PMS API using the caller’s phone number mid-conversation Option 1 feels simpler but is a data protection minefield Option 2 is what I am hoping will work but potentially adds latency and another point of failure. Has anyone actually shipped this? What approach did you go with and what PMS were you working with? Would love to hear what worked (and what didn’t)

So my ai is in Dutch, and whenever a user spells their email adres to the ai, and the ai tries to spell it back, 99% of the time the ai is inaccurate, same for street names. Also when the ai is spelling it out, it says the letters of the alphabeth all so vague, it only says half of the letter, thats the best how i can explain. And when it has to pronounce numbers it also says them so vague i cant understand what it says, i prompted the ai to say everything correct but it cant manage.

long dev cycle but we've rebuild the platform from the ground up along compliance audits. its a lot. but, something we have put into product for managed clients is the production version of flow builder for voice and chat. I know a lot of people have wanted this for some time, we have to.