AI-native Personal Financial Advisor (RAG on Azure)
FinAD using Azure Open AI by Vishal Anand.
Section 1: Conversation with FinAD
Section 2: Azure OpenAI Architecture for FinAD
Section 3: Next Steps
Section 1: Conversation with FinAD
After last night’s storm, a beautiful morning by the sea side town in Dublin, on a Saturday — could not be a more perfect day between me and Azure.
I decided while having a cup of tea that AI must start at home for everyone in this business, specially at the home of a technologist who has tremendous passion for technology to influence business outcomes.
So, I decided a use case where I would train Azure Open AI with my financial transactions based data. I love to start with the hard part first, hence a complex use case, which always gives me deeper knowledge on the subject matter for wisdom.
Was I worried about my data ? Honeslty, I trusted Microsoft’s commitments in this regard. Read the following:
Ok, back to the use case..I spent approximately two hours in building/deploying the whole architecture required on Azure cloud, integrated various components of cognitive and storage services, deployed the Open AI model with gpt35turbo-16k, sanitised the data, injected the data, engineered a few prompts and so on.
Note that the base model here was gpt35turbo-16k, on top of which I trained the model with my dataset.
I had FinAD, my personal financial advisor up and working. Let me show you my conversation with FinAd.
-> -> Fact 1: It controlled itself within the data boundaries and hence the desired completion.
It passed my very first test by not answering who was the prime minister of Ireland in 2020 (it was not in the data). At the same time, it answered which bank I use.
This established my trust to start with. Now something more interesting happened in Figure 2.
When I started chatting, it greeted me and rightfully assumed the responsibility of my banking transactions and financial needs.
-> -> Fact 2: This was the magic of System message I configured as a prompt which shaped its very behaviour.
-> -> Fact 3: It provides Explainability with Source for each completion/outcome. It shows citations with actual data sources and the portions of data. As you can see “references in chat windows at the bottom left corner”.
-> -> Fact 4: It could provide me with right contact numbers when I expressed my concern and asked for the customer service number — quite handy.
-> -> Fact 5: It could not provide me with more Starbucks and Tesco transactions which were available in the data. This requires more investigation (which I will absolutely do).
Section 2: Azure Open AI Architecture for FinAD
Ok, I literally breath and live complex architectures everyday. I am not going to write much about it on a weekend, apart from the fact that it is all Azure-native, Azure Open AI native and hence a complete AI-native solution.
Section 3: Next Steps
I will further train it with more data, more integrations for providing more insights and recommendations. Sky is the limit for such use cases.
Imagine UtilityAD if I train it with my utility consumption data. Imagine InsuranceAD if I train it with my all insurances data. And, so on…..
Note: A lot of the architectural features used here are still in preview phase by Microsoft. Watch this space.
My intention here was to share with audience a glimpse of the power of Generative AI and what all can be achieved for making our daily lives easier.
Disclaimer: Views and opinions are my very own and really personal ones.
