I am making something for the community. Need Feedback

Model loaded: Qwen-3 1.7B 4bit What I am trying to do in layman terms: I want to create a close to Perplexity experience with your locally downloaded GGUF. Here is one example of the Deep Search feature(I've cut nearly 30 seconds of the video while it was searching). So far I've implemented complex pipelines and steps of the model searching with memory and none of your data goes anywhere(no api calls, search is implemented using searxng) How are the results for a 1.7b model? would you use something like this? I will be adding more features in the coming time and will make this 100% open source once it reaches zero to one. What features would make you switch to this instead of whatever you are currently using.