Working on https://practicecallai.com/ - simple saas that lets users run practice calls / role play against a custom AI partner. Goal is to make it the easiest to use & fastest to get started with in the market.
It’s been a fun, practical way to continuously evaluate the latest models two ways - via coding assistance & swapping between models to power the conversational AI voice partner. I’ve been trying to add one big new feature each time the model generation updates.
The next thing I want to add is a self improving feedback loop where it uses user ratings of the calls & evaluations to refine the prompts that generate them.
I was getting stuck in some manager <> direct report situations and wanted to figure out a way to practice tough conversations, so I started putting together https://practicecallai.com/.
You can choose from common difficult scenarios or create a custom one and get connected with a conversational AI voice agent to practice. Started out with just manager use cases but now testing out sales calls, interviews, pitches, etc.
Mostly built on Replit (would recommend - is amazing & keeps getting better) and relying on a mix of ElevenLabs / OpenAI / Anthropic for the voice & LLM tech.
I've been working on an iOS app called snapmail: lets you take any photo on your phone, filter / sticker it, and send it as a 4x6 postcard anywhere in the U.S. for $2.99. Fun way to stay in touch with friends & fam. One of a few no-account-needed, single use iOS apps I've been working on.
https://www.usertesting.com - you can upload screenshots / prototypes / the actual webpages (if they exist) and have users go through a series of questions. You can also set it to scramble the order to see how that (if at all) changes preferences.
As other folks called out, this can give you a sense of customer comprehension / stated preferences & reasons why, but may not give you the level of confidence a true A/B test would (launching both iterations & splitting traffic amongst & measuring conversion difference)
Which market are you in and who did you end up getting in touch with in support? Did you try the Safety Incident Support Line? At least in the US that is 1 (800) 285-6172.
Without getting into specifics they absolutely do have a way to track down the driver regardless of the state of the trip / who cancelled.
Yes I specifically called that line. They informed me that they had no ability to see driver canceled rides. They could only see the same rides I have in my history. They were sympathetic and took my complaint but ultimately it ended up nowhere. The same Uber support black hole of nothingness. No response, no follow-up nothing.
Pretty much this driver exploit can be executed with impunity and Uber does not care and there is no way for customers to report it.
I have a good friend who runs http://barnonetrivia.com - they have professional hosts who lead ~hour long live zoom trivia games. Questions are creative and original. Can't recommend enough - very fun!
I know Nick, Bob, and Strader from early Uber Boston days - they are great people and very hard workers. I'm really happy to see ex-Uber folks going after an opportunity to strengthen local communities vs. so many others who tried to hop on the next scooter / marketplace / fintech 'rocketship'
I don't know enough about the investment model they support, but any company that is trying to help grow important community businesses like Great Scott in Allston deserves attention.
https://www.higifter.com/
It's an SMS-based chatbot that helps you remember special occasions (birthday, anniversary, mother's day), recommends gifts, and purchases them for you automatically.
Working on this with my girlfriend, mostly for fun and to learn something new. We're using Webflow for the site, Airtable as the backend, Zap to help stitch a few parts together, and Twilio for SMS support.
It’s been a fun, practical way to continuously evaluate the latest models two ways - via coding assistance & swapping between models to power the conversational AI voice partner. I’ve been trying to add one big new feature each time the model generation updates.
The next thing I want to add is a self improving feedback loop where it uses user ratings of the calls & evaluations to refine the prompts that generate them.
Plus it has a few real customers which is sweet!