This is great, will definitely give it a try. We had such bad experience with Wit.ai and Api.ai, (still many bugs) and I think it's not worth yet building your company/product based on it.
Hello PH!
As much as we love third party NLP APIs (wit.ai, LUIS, api.ai) for making prototypes, we found that when your bot gets some traction you inevitably want to become independent. Why should you use rasa NLU?
- Own your data, don't hand everything over to Facebook / Google / Microsoft
- Faster : Runs locally so no network overhead
- Customisable : get better performance by tweaking models for your application.
It doesn't make sense for every bot team to re-build their own NLP, so we thought we'd open source ours so we can all collaborate instead.
This is our first step in making the bot community more open & collaborative, and we're extremely excited about the enthusiastic response we've received so far π
Happy to answer any questions :)
rasa was easy to train and customize to our use cases. We really like the extra speed gain without the roundtrip to an external provider, plus keeping our training data under our control.
Great product overall!
Like the project. How does mitie works in comparison with spacy/sklearn backend, do you have any experiences or recommendations for specific use cases?
Looks great Alana, as was about to jump seriously into WIT. Thanks for doing that, however can you give your measure on time spent to design dialogs, suppose you have nothing like UI tool to help in doing so? (just for lazy people to write a code)
Wow, really nice to see this on PH. I recently started building an intelligent agent to run my home automation. Am using Wit.ai, but seriously considering changing to Rasa. Well done @tombocklisch@alexweidauer & @alanmnichol
@alanmnichol I was looking for such solution. Great work and commendable step to open source it. can you tell us more about your thought process to take the decision to open source it, do you have any plan to support it's further development?
@pradeep_io we organise a bot meetup in Berlin, so we've spoken to lots and lots of bot developers. We saw that the teams who were a bit further in terms of traction were shifting away from 3rd party APIs, and we just thought it would make more sense to collaborate on a library rather than have everyone reinvent the same things :)