STEVE   KING

CONVERSATIONAL   DESIGN   EXAMPLES

For me the whole AI / NLP / ML thing started 20+ years ago with another acronym, good ole CRM, and scripted customer flows:

Flash forward 20 years and I'm still working on conversational apps, but now the stakes are higher because everyone wants to see what AI brings to the commercial software mix:

Going back to the deep dark (nearly) pre-ML days... I won a CRM design competition with my idea for a conversational "social CRM":

Here's the award... it convinced me that conversational apps are cool and here to stay:

Here's a social graph network diagram that's automatically generated by conversations in the social CRM app:

More recently I was principal prototyper for this amazing AI driven customer support bot that's used by major travel, hotel and retail companies today:

Here's the dashboard for that travel/hotel bot:

I've worked on many clinical apps that are data rich and by nature conversational in the way data flows from handhelds to reporting screens and back again:

A design I did for an app that creates conversations between hospital staff and the family of patients who are undergoing surgery (ylo Care Circles):

The family user interface for the Care Circle hospital app:

Tablet, Android, Apple, desktop, laptop, large screens... all good vehicles for conversations:

A portable printer used by clinicians on the hospital floor... wirelessly tethered to the mobile app. The software has to fit seamlessly into the conversation the clinician is having with the patient:

The console where mobile clinical apps and incoming patient notes are monitored:

MY DESIGN AND PROCESS TOOLS

AI is great by itself for a lot of generative, creative, planning, reporting applications, but for many operational and transactional apps, AI is best when integrated with proven ML, NLP, IR, DBMS, OLTP, etc. disciplines

I created an innovative conversational order flow app for a large manufacturing supply chain in the commercial printing industry:

Because the app crosses suppliers, distributors and manufacturers... lots of flow diagraming and visual requirements analysis was needed:

The key to conversational design is understanding dialog patterns, so flow charts, entity maps and state diagrams are always helpful:

Creative approaches to documentation always help shared understanding within the dev team:

Stanford NLP / Natural Language Understanding

Stanford Conference Paper

 

Here's a NLU design studio that makes efficient AI chatbot design possible (Voiceflow, which is similar to Google Dialogflow in functionality):

It turns out AI large language models and RAG are very susceptible to 'initial conditions' (settings, prompting, parameter tuning, in context content, etc.) Consequently rigorous testing must be baked into the infrastructure where AI apps operate:

Fine-tuning, in-context learning, RAG, prompt engineering... al important methods but for apps requiring high levels of deterministic accuracy and consistency, there is no magic bullet. Here's some key AI/ML/NLP building blocks that can wrap-around AI and make it more viable for business apps:

Some of my favorite libraries and their developer support on Github:

Google is a key player in the NLU/NLP/IVR space. Here's their stack with Diagflow and Vertex AI on the top tier:

Using Scikit-Learn and related ML libraries to cluseter 5,000 hotel guest chatbot queries. Clusters help define topics, intents, entities and NLU flow logic:

Using Python SpaCy NLP tools to examine the hotel chatbot corpus and show diffusion of key utterances like 'parking,' 'restaurant,' 'hotel room' ...which drives customer intent in an NLU application:

Classification sandbox in Jupyter with K-nearest-neighbor, regression, naive bayes, support vector machine, and neural net competing for best precision, recall, F1 scores:

VIDEO   CLIPS

 

FINAL   THOUGHTS

Some folks in the industry caution that giving product managers chatbots is like giving children loaded guns... i.e., recipe for disaster:

But we have some powerhouse libraries and frameworks at our fingertips... so let's put heads together and think about how to make this work !

LinkedIn   Skills   Profile

LinkedIn bio: linkedin.com/in/stevebio

NEXT   STEPS

My AI/ML Portfolio

My Group Intelligence Research

My UX Portfolio

My Content and Writing

Home

Let's brainstorm exponential opportunities together:

 

Back to top