Circuit GPT

Some Brainstorming before we took the train to OpinHacks
Bhubaneshwar Unit-8 Substation
We went to substation for checking out how equipment's testing works
Here an engineer was giving us a demo on how they perform IR test on transformer
Group Photo with our team and substation engineers

Inspiration

We've been dabbling around electrical engineering space for a while, coupled with discussions with friends in this field and that revealed a significant technology gap. Inspired to make engineers' lives easier, we wanted to take some action to bring smart technology to core electrical engineering areas.

What it does

Circuit GPT is an AI assistant for streamlined substation tasks, from equipment maintenance to data retrieval to breakdown resolution and more. Designed for substation engineers, it saves time and enhances focus on high-priority matters.

How we built it

For the first half or the initial version of the product, we went on and fine tunned GPT-3.5 turbo and trained to meet our specific use case. We collected the data from few substations and trained the model for giving specific answers. We also went on and trained the model on DGA test result which is used to predict the transformer health condition. Now with this in hand, Circuit GPT can predict Transformer health condition (we plan to do it for most of the equipment's on substation). We also went on and create a high fed prototype for v2. It involves giving specific ID to each substation, allowing engineers to upload data set and asking queries regarding that data set and more

Challenges we ran into

The major challenge revolved around acquiring, cleaning, and structuring the data effectively into query types, inputs, and outputs. Inadequate or low-quality historical data also posed a hurdle in enabling the Assistant to give accurate insights.

Accomplishments that we're proud of

We've made some great progress and I'm proud of our efforts. Before reaching here, we conducted on-site research at substations , built the UI, designed prototypes, and cleaned data for transformer health prediction during this hackathon. In just two nights, our team achieved a lot...

What we learned

There are so many things, from data cleaning and formatting to diving into fields like electrical engineering, it's been a rewarding experience. Despite our CS background, we can now confidently say that our involvement in this project has expanded our skills into these core domains.

Can't miss out on how much help our mentors have given us. They've provided invaluable guidance on tackling various challenges. We're eager to implement many of their suggestions in our v2.

What's next for Circuit GPT

Well there are so many things, we want to try out here but I'll list out some of high priority ones

Enhancing Backend: We're focusing on improving the backend infrastructure to improve the data uploading process further.
Leveraging Document AI: We're integrating Document AI to simplify the uploading of testing result pictures, transforming them into actionable data effortlessly for substation engineers.
UX Refinement: We'll also be refining our user interface, enhancing the overall user experience, and creating a more intuitive flow for substation engineers.
Equipment Health Prediction: We'll be expanding more on health prediction of different equipment's in substation, most importantly transformer.