Building an AI-trade-Assistance on very confidential private data

AI Discussions
1

Hi reader,

So, I am building an AI trade assistance on very sensitive and confidential user data, hence the need to ensure EVERYTHING is done locally without calling any external APIs.

The AI agent should be able to understand the user’s queries and perform the desired actions. For example,

user1: What is my account balance?
operation: the AI agent should fetch and return the account balance of user1 from the data-balance table, where userid = user1 as AC BAL.
response: Thank you for asking for your account balance. As of today <today’s date>, your account balance is .

user4: Kindly place a trade order for me in the next 3 days for XYZ
operation: the AI agent should trigger the API for placing an order, supplying the specified date as today's date + 3 days. If the returned response is completed, return it as successful.
response: Thank you for placing that order. I have placed the order, and a confirmation has been sent to your email.

Any idea on how to approach this, please? This is a real-life project I am working on, and the time to deliver it is in 7 days. I have access to Google Colab Pro for all my training and testing.

What I have in mind is to LOCALLY fine-tune an open-source LLM (maybe from Huggingface) and save the fine-tuned LLM locally, to be then used for inferencing.