arrow_back How to build scalable and robust data pipeline iteratively.
Talk Less, Chat More
Submitted by Ashutosh (@ashutrv) on Friday, 2 June 2017
Full talk for data engineering track
Conversational interfaces are the new channels coming up for business. These channels are new for both users and businesses. For a business it’s a new kind of user behaviour they have to understand! This new behaviour generates a completely new kind of data.
This data will provide a completely new information or improve on existing information, a business has about its user. These channels are intimate and personal. This talk will try to give a basic understanding of this new channel from data, analytics and business point of view. To list a few:
- What new platforms are doing.
- How’s the technology solving it.
- How to manage the data which is being generated
Conversational Interfaces ( 10 min )
- Introduction of the paradigm.
- Why is it important to understand this, for business and techies both.
- What are the use cases?
- New Channel - New kind of data.
- How to look for information in this data?
- What problems can this data solve/improve?
Platforms ( 15 min )
- Facebook Messenger as a platform.
- Architecture overview.
- kick start a FB bot in 2 mins.
- challenges and workarounds.
- Limitations of current platforms.
- Future Roadmap of platforms.
How Far is AI ? ( 10 min )
- Retrieval based models.
- Generative models ( a demo of pre-trained model )
- Markov chain based model.
- How to workout the context in conversation?
- How good is AI as a service from wit.ai, luis.ai, api.ai?
Where Are We Heading? ( 5 min )
- How is the future look like optimism or whishful thinking?
- How are the industries looking at the bot channels?
Ashutosh is co-founder of PerfectPi, A platform to help brands to create conversations on FB Messenger. Ashutosh is a machine learning and NLP scientist. At PerfectPi he has built the platform from ground up and scaled it to handle millions of users.
There were so many learnings in this journey of building PerfectPi and want to share the experiences/ challenges this field faces from technology, data and business point of view.