ChemoChat
Filter live stream chats: only interesting messages & info
Inbound ad campaign has started and with it the few first users. Still early to tell whats the conversion rate.
A lot of tweaks about how to get rid of as many cookies/localstorage as possible and avoid processing PII. Privacy and Terms of Usage page were updated.
It still doesn't work 100% of the time because the underlying library isn't official and relies on third-party service that sometimes goes down. But it is up a good percentage of time so it is available to use.
It's funny how much more useful ChemoChat is there because annoying text isn't filtered so often as it happens on Twitch.
AWS has become Chemo.chat definitive home at https://chemo.chat
As always, still a few kinks to work out but it's ready for anyone to pick it up. I'm now entering a phase where I'll invest much more time in initial marketing & community research.
Twitch is working ok, TikTok was developed but the working library is too unpredictable to use, so I'll have to wait a little more maturity from the ecosystem there. YouTube can't be used at the moment until I get a few streamers willing to be alpha testers.
So AWS went thru for us and has given us AWS Activate credits, which would last a good few months, maybe an entire year with our current usage. Turns out it was a good move from Digital Ocean to AWS, which I'm much more familiar with their quirks.
Next level regarding infra would be to use an entire fleet of spot instances to works sockets & message handling but that is on the future as it is overengineering at the moment.
After all this time, ChemoChat is finally at a demonstrable state, where its UI, UX and sentiment analysis are solid for a first version.
At this point gauging super early interest is the goal, so no Pricing pages, even login is optional; from a tech perspective I only want to measure how much I can stretch a single server.
After a few more stuff is ironed out, I'll declare "open beta"
Bullet points until beta:
- Increase security;
- Review accessibility by using JAWS.
- Cheers & bits overhaul.
- Little more data training.
The last major feature for Twitch channels is in progress: show a bar regarding what kind of sentiment is being told by user's messages. I've finished the alpha version but more work is needed until it can faithfully reproduce what's being said. A lot of research had to be done first on the "dialect" usually spoken in live chat, because of course formal language isn't suited for it so classic approaches don't result in useful info.
A few other advances were made as well:
- OSB and Streamlabs plugin were tested and the frontend works with little adaptation;
- Auth0 and Stripe integration are underway.
These past few weeks I've been working again at the project, and some critical milestones were achieved:
- Backend is scalable towards running channels concurrently;
- Cancer analysis is working;
- Frontend has a good looking way to show chat usage;
With those steps in place, the project is close to shipping Beta as it ever was.
-
1
Hey Fernando, nice work on achieving those milestones! Was going though your posts, didn't find any content around general research. Just curious have you conducted any research? Just intrigued to see how you found that process?
-
1
I did some research, yeah, mostly regarding sentiment analysis. That's a good idea for a longer post.
-
1
Did you validate your idea at all? I'm going to be upfront and honest, we're building a product that aims to help founders conduct user-research and validate their ideas. Just trying to understand what/if you had any issues at all during this process.
-
1
Oh you meant customer research. Yeah, when the idea struck and I did the first POC work, I reached out to a streamer friend about the value propostion, which he then asked a few streamer friends about it too. They were definetely interested in something that could help make better sense of what's going on in the chat.
I do plan to run a more in-depth, 1 on 1 feedback cycle when the first version is out, so I can better understand how to adapt the UI for how a streamer works. But as far as validating the idea, showing the BMC at the beginning help me understand there was a potential product there.
-
-
-
Wow, a year has passed since I've been working part time at this idea.
As it is, backend has a proposition value, but some features are still lacking and as it is, chat screen is more at a proof-of-idea stage.
Frontend design is still kinda painful with React + Websockets. It has a lot of bugs.
I've recently came across PETAL stack and I think it suits perfectly for the job and could accelerate overall deliveries. If I can migrate from NestJS to Phoenix (most of the code has unit test already), I think it would lead to fast benefits in terms of delivering with less bugs.
As of now, the stack would be NNARW (Nest+Node+AntD+React+Websockets), so PETAL makes lot of sense.
-
1
I totally dropped any major changes to the stack. It is working fine as of now and I see no problem shipping this way.
I used my free time to test hypothesis on what kind of tech stack & solutions I'd need to create an app like this.
It's aimed at Twitch chat but it should be extensible to YouTube in the future.
Twitch fully support IRC as a way to subscribe to channels, so that'd be my primary gate to create a client that would republish filtered messages to our users.
Streamers can't possibly read anything from live chat.
What if we apply AI spam techniques & UX design principles to create a heavily opionated chat bot to help streamers interact with their audience?
The name is a pun with a common complaint in live chat ("this chat is cancer/toxic")
Comment