How Bimpe AI Handles User Privacy
At Bimpe AI, we are a privacy-first company, and we take our customers’ data with care because we understand how important privacy is.
We have implemented data governance and protection principles into building Bimpe from scratch and have detailed how we use the little data we collect in our terms of use.
What data do we collect?
We collect your name, email, and profile display picture from Google to help us identify you and make Bimpe personalised to your profile. An early feature one of our customers wanted was that Bimpe should be able to know their name, and that’s why we collect your name alongside your email. We are building guardrails to anonymise your identification data on our backend.
We log your chat on Bimpe AI web on your permission
Some users want us to keep their older chats on, while others want us to delete them immediately. We solved this by giving our users the preference to pick whether to keep their chats or delete their chats. This is deleted automatically from our logs.
Does Bimpe AI log WhatsApp chats?
No, we selected WhatsApp first because of its end-to-end encryption technology, which makes it almost impossible to log chats. This means all chats with Bimpe are saved locally on your device.
Does Bimpe AI share my message with LLM models?
We currently use two LLM models to power the Bimpe Engine, Llama 2 and the Open AI 3.5 turbo. Meta Llama2 is open-source, so we have to train the model and host on our private GPUs, creating a closed-loop system. However, with the Open AI model, we connect with their APIs to connect their tokens. Whether we use Llama2 or 3.5 turbo, we don’t store your conversations with Bimpe on any of our LLM models.
LLM models are generative models that generate responses based on prompts. It can also be referred to as a transformer model. What this does is pick key prompts and generate responses for the result. We use LLMs via an API or a wrapper, the LLM doesn’t understand what is being sent or received. Your personal information is relayed to and from this generative model and programmed to respond to you in a specific way. Think of this as a radio signal feeding you sound waves. When you tune to the right channel, your prompts are the tuning.
We work on how to improve our privacy framework every day and we are open to taking your suggestions and questions here — hello@bimpe.ai
