How to Deliver AI Prototypes and Proofs of Concept - The Key Questions Answered

At Affino our initial AI Prototypes were years in the making. There were a number of key reasons for this, not least the lack of ’good enough; commercial grade AI APIs and services. What was clear though was that AI would reach an inflection point at which time it became an overwhelming imperative, and that time has arrived.

 

Affino has been looking to deliver our SaaS services with AI at its core, and we long since migrated to support Python as one of our key languages and increasingly developed and deployed our services via serverless solutions. Moving to Python and AWS has been key to our rapid adoption of AI as it is the preferred AI programming language.

 

Everything changed with AI for us when ChatGPT and MidJourney hit major quality of service milestones in November 2022. It was also the start of the rollout of enterprise grade APIs for AI services.

 

AI service architectures and data services also reached a level of maturity where it meant they could deliver on the quality metrics we had set out as being the prerequisites for us to roll out AI into the core of Affino’s service architecture.

 

We spent the first six months of 2023 developing a series or AI Prototypes, in what was in truth a fraught process. The single biggest challenge with rolling out AI Services is that they require a mental shift. AI’s simply do not work the same way as traditional services you might have developed. They are something new!

 

AI related tech is also evolving at a phenomenal pace. In fact one of the biggest hindrances to AI adoption is that the pace is faster than any AI course, help guide, or even YouTube video creator can keep up with. It means that you need to learn and adopt AI at a more fundamental level than most technologies which evolve more slowly. It took a while for us in the Affino AI team to realise just how fast this evolution was happening.

 

AI’s Instantiate Abilities, they are not programmed in the traditional manner of software development.

 

At this point everything has clicked, and we have already evolved a third generation AI infrastructure for the Affino SaaS service. We have also launched three in-house AI Prototypes which have gone through multiple stages of evolution and are adding value daily to Affino.

 

The initial three Prototypes we developed are: the Affino Spec Bot (fed with 16,000 Affino 8 specifications; the Affino Release Bot populated with all the Affino release notes; and the one we will be launching is a customer facing public beta this week, namely the Affino Support Bot.

 

It is simply remarkable how effective these have become for answering Affino feature and functionality questions and in providing meaningful responses. They are having a significant time and cost saving impact, and indeed the quality of the content they are producing is in many cases at least as good as if the relevant Affino team expert was providing the answers.

 

To deliver on Affino’s AI Prototypes and now our customer facing AI services, we have rolled out what we believe to be a best of breed framework and approach, which can take any content, originating on Affino or elsewhere, and deliver insightful services using the content via AI services.

 

Here are some of the key questions we have had to answer along the way, and the key takeaways we have for each one.

 

What is the best way to Get Started with AI as an Organisation?

The best way to get meaningfully started with AI for us has been to identify what are the quickest ways we could start to empower the team.

 

We identified three areas which would immediately help speed up our support, discovery and development services, namely taking all our key specifications, help guides, and various support FAQ’s and converting them into conversational bots. It took a few prototypes to get them to the stage where they provided meaningful answers to complex questions, but equally we feel they are already providing an essential service used across the organisation.

 

What are AI Prototypes and Proofs of Concept?

AI Prototypes are early versions of the AI services you might look to roll out. They are highly experimental and leverage your content and data to deliver meaningful services to your organisation and wider community. Proofs of Concept are the next step on from Prototypes, whereby the prototypes are developed into complete services.

 

These services could either be focused on delivering insights to your core team, or taken further could be closed trials, or even new startup brands for exploring new services and revenue models.

 

Why Should we create AI Prototypes for our organisation?

The simple answer is that AI will change a huge swathe of how we work and are entertained over the coming years. AI’s can already deliver an exception range of innovative services. Converting content into supportive assistants, which are able to both consume and deliver experiences across multiple media formats and user interfaces.

 

Is now a great time to get started with AI Prototyping?

Yes. The AI platforms, API’s, and associated services are evolving so rapidly and maturing quickly. It means that organisations who are not actively prototyping services soon will be at a significant disadvantage which will simply grow larger by the day.

 

How can we get started with AI Prototypes?

You can either develop an in-house AI team with technical, development, content and commercial leads, or you can work with an agency or partner service provider to roll out the AI Prototypes.

 

How can we identify good candidates for AI Prototypes?

Look to see where you have great content, and / or meaningful challenges which can be addressed by AI. These could either be there to provide services in-house or to your customer base.

 

What are some of the key things we should look to achieve?

The most obvious service which AI can provide is for support and service assistants. Taking your existing documentation and transforming how it can quickly and meaningfully provide answers, insight and advice. AI’s are also exceptionally good at format transformation. Taking audio content and converting it through to text, and then on to video. AI’s can also provide considerable insight and data manipulation capabilities quickly and cost effectively.

 

What might an AI Achitecture look like?

There are eight parts to any AI archictecture: Content / Data origination / ingestion engine, e.g. CMS / CRM; content conversion to AI data formats, i.e. Embeddings; AI content storage, e.g. Vector Databases; LLMs (Large Language Models) to interpret the data; User Experience suite to manage the AI prompts, conversations and guardrails; the AI teaching / feedback engine; and finally the AI Analytics, Dashboards and Reporting.

 

These are in essence the foundations on top of which you build your in-house and commercial services. To deliver fully you need to plug these into your overall service architecture and extend with subscription services, ads, ecommerce and other commercial engines to ensure your best return on investment. 

 

What are the Key Components of an AI Proof of Concept?

The two most important elements are: What is the output you’re looking for? Is it a commercial service you are looking to monetise through subscriptions or is it a new in-house AI service you are keen to implement to gain efficiencies and further insight. Once you have identified the outputs, check to see if you already have the content and data available, and if not how it can be sourced.

 

Once you have these two then you will either need to simply engage an AI Prototyping and Proof of Concept service provider such as Affino, or you will need to develop the extensive skills and architecture in-house.

 

How do we Create and Refine the content for the AI?

This is one of the reasons your organisation should explore AI services at the earliest opportunity. You are likely not working with your data and content in a way that is optimised for delivering AI services in the future. AI’s need content originated and prepared in specific formats to be the most effective.

 

The fastest way to figure this out is simply to start prototyping your own AI’s. Quick tips though include: Keep your content in a single format, e.g. text, images, videos or audio. Mixing them makes it far harder to use them as sources of AI service. Make sure you provide the full context within each piece of content or set of data, e.g. have a table title and description immediately above any data table. Possibly the most important element is to set up effective debugging, training and reporting systems so that you can see exactly where the AI’s originated their insight from so that you can further refine this.

 

In very short order we have evolved our AI content strategy in Affino rapidly to respond to these evolving requirements. All content is near realtime fed into AI’s and the team can immediately assess the effectiveness of the content in educating the respective AI service.

 

What are the Best AI Prompts to Use?

This is a great question and depends very much on the content you have and the type of service you are looking to provide. In practice you evolve both the content, the embedding (encoding) of the content and the prompts alongside each other to ensure the most effective outputs.

 

Are all LLMs (Large Language Models) the same?

No, we have found vast differences in differing LLMs. The more modern, the more effective they are as a whole. That said, you should look for the right LLMs in terms of efficiency, cost, and indeed capabilities to match the services you are looking to deliver on.

 

How fast is AI Evolving? Can we keep up?

Right now the jury is out on if we can keep up with the evolutionary speed of AI’s. It is possible that we will hand over the development of AI’s to AI’s themselves and they will rapidly evolve beyond our performance capabilities.

 

There is also a chance that right now we are at peak velocity with how fast AI services are evolving and we will start to see things stabilise over the coming weeks and months. Right now though there does feel like an ever accelerating  cycle happening which may well lead to something completely new for us all.

 

How can Affino Help?

Affino provide a suite of AI related services from advisory through to prototyping, proof of concepts, and indeed working with you to roll out entirely new commercially driven AI ecosystems.

 

We already have a solid AI architecture and scalable infrastructure in place to roll out AI’s on. We also have the expertise to ensure that your AI projects are delivered from start through to completion, and with the ongoing service delivery for years to come.

 

If you wish to explore what AI can do for your organisation simply reach out to us and we will set up a call, Zoom or Google Meet to move things forward.