When Lobo invited me to a meeting with Dhaval from DanaMojo and Noshir from CAP , I didn’t expect to be part of such an animated conversation. The two obviously knew each other for a long time and the idea of a legal BOT for NGOs had been playing through their minds for a long time before they approached Glific.
In fact Lobo and I seemed to be just listening into their conversation while they worked out how they would go about things. Dhaval quickly offered to handle opening of all the accounts needed to get the NLP BOT runnning – the Facebook, Gupshup and Google Cloud Platform accounts not to mention the Glific SaaS instance. Noshir would provide his expertise.
In fact Noshir already had a significantly busy WhatsApp group where he would answer legal questions NGOs had on CSR fund management, managing donations from another country , taxes – all the stuff that an NGO needs to figure out but has no expertise in.
By and by the accounts were set up, Noshir provided the FAQs and I set out to work configuring our BOT- “DanaVidya”. Google DialgFlow has provided a new beta feature called “Knowledge Base” . I thought I would give that a go as opposed to build a BOT with intents from scratch – and wasn’t I glad I did! With the knowledge base feature creating a BOT from a set of FAQs can be as easy as uploading the FAQs document into the knowledge base. The default implementation does such a great job of matching questions to the right answer ! It manages to do so even for a wide variety of ways in which you frame the same question. You don’t have to go through any of the painful business of developing “training phases” for your BOT. You are just done !
When I circled back with Dhaval and showed him a demo, he was quite excited about it and quickly went on to thinking of the next steps. What if the user asks a question outside of the scope of the FAQ? How will the BOT respond ? How can a staff member intervene ? How do we analyze how many questions get answered automatically and how many needed human intervention ? I explained about the “fallback intent” and how we can configure a notification in Glific to a staff member in case the BOT does not find a response to a user’s question. We can also set up reports using a combination of BigQuery and Google Data Studio for him to understand how well his BOT is doing and where manual intervention is needed. This can provide opportunities to further improve the FAQs.
All in all, it was a very satisfactory exercise. I was amazed at how easy it was to adopt the technology and how it freed us to think of just our problem and not waste any energy wrestling with technology! Hats off to Google though for packaging what they do so well and making it so accessible to businesses at so little cost !
[…] I was able to help people thinking in terms of how NLP can be used seeing the TAP and DanaVidya use […]