Intercom was shipping and experimenting with AI enabled products since 2017. This episode is for those interested in hearing about the evolution of AI models, the customer results, and how these models can be applied in production.
Fergal Reid is Vice President of AI at Intercom. He started at Intercom nearly 7 years ago to help the company begin delivering products leveraging AI. In this episode, we cover how ML & AI has evolved since 2009 and why the reasoning engines that exist in models today are a substantial improvement. We dive into Intercom’s AI-enabled products starting in 2017 to now Fin (Intercom’s AI chatbot) in 2023. Fergal dives into the technical considerations, how the team delivers usable AI enabled software, dealing with hallucinations and the tangible outcomes that Fin has enabled for Intercom’s customers. We discuss latency tradeoffs, using multiple models, building MVPs, and much more. Teams at any stage of their AI journey will benefit from this episode.
Where to Find Fergal:
* LinkedIn: https://www.linkedin.com/in/fergalreid/
* Twitter: https://twitter.com/fergal_reid
Where to Find Shomik:
* Twitter: https://twitter.com/shomikghosh21
* LinkedIn: https://www.linkedin.com/in/shomik-ghosh-a5a71319/
* Newsletter: https://www.shomik.substack.com
* Podcast: Apple Podcasts, Spotify, YouTube.
In this episode, we cover:
(00:49) - Fergal’s Background
(03:23) - Why This Wave of AI is Capturing So Much Attention (pun intended)
(05:51) - Capabilities of Any One Model Across Multiple Domains
(08:04) - Evolution of Intercom’s AI Enabled ChatBots From 2017 On
(15:07) - Managing Latency & Performance Tradeoffs Amongst Models
(17:11) - RAG & Other Considerations Accounted for when Shipping Products
(19:43) - How Much Prompting is Needed for What Use Cases
(20:41) - Routing Context Between Multiple Models in an Application
(23:14) - Updating Context in Vector Embeddings
(24:58) - Best Practices for LLM Data Access
(27:04) - Outcomes from Customers Using Fin
(29:08) - Customer Response to Hallucinations
(32:09) - Pricing Based on Successful Resolution
(35:28) - Advice for Technical Teams Experimenting with LLMs
(38:20) - LLMs as Reasoning Engines vs Fuzzy Databases
How to Subscribe:
Available on Apple Podcasts, Spotify, YouTube.
This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit
shomik.substack.com
