BugSensei

Inspiration

Running into issues with our laptops, and other personal devices has become part of our everyday lives, but resolving those issues has become quite tiresome and expensive even. Most software issues could be solved through simple google searches or through extensive ones where it involves us visiting several user forums. But after identifying the problem, solving it is another thing, that is why BugSensei also has proposed a solution to this one as well.

What it does

BugSensei is a dynamic RAG conversational assistant, it listens to your issues and gives you the best possible methods to solve them. But that is not even the best part, based on the responses generated, it also generates powershell script to resolve those issues in one go, only if it is not a hardware issue or requires manual GUI toggling.

How we built it

We ensured BugSensei has a wide arsenal of bots or agents that could retrieve data from the net, which served a crucial component in the responses being specific to the user issue. With Snowflake's help, we were able to warehouse these data and effectively retrieve them. But Snowflake also provided other compute services as well, such as Embedding Model, LLM Inference, Similarity Search, which we used quite thoroughly in our pipeline. Once we had the necessary data in hand, through a decision agent, we selected the best response based off relevance and generated user-friendly responses. And the decision agent also sends the top 3 responses or contextual information that is staged to our database, this continues to ensure to enrich our database with rich data, thereby reducing dependence over web retrieval over time.

Challenges we ran into

We ran into multiple issues, some that continue to persist even now. First off, the sleek UI which we developed for BugSensei wasn't implemented due to CSS issues, it ran quite well on local host but when deployed off Streamlit Community Cloud, the CSS alignments and stylings were considerably well off, which forced us to adopt a basic chat UI for BugSensei

For this project, we fine tuned a Mistral 7b v0.3 model for generating powershell scripts(for now), but unfortunately we were unable to host off the compute service at Snowflake with the reason being that ours was a trial account.