• 

  LoGPT

LoGPT

Checking your logs made easy.

Inspiration

Going through the log files is a tiresome process as most of the log files are very long and complicated. We present LoGPT, a way to communicate with your logs through a chatbot that highlights and summarizes parts of logs that might be of interest to the user.

What it does

LoGPT computes clusters and extracts embeddings from the log file (or stream) through a Transformers-backed language model. Using either our mobile or web application, it takes a textual input prompt from the user and provides this to the model to extract a query embedding. Then, the cosine similarity between the query embedding and the log embeddings is computed. The log lines that are closest to the user query, together with a GPT3-formed summary are returned to the user through the interface.

How we built it

We used Next.js, Node.js and Firebase to build a full-stack web application, and Mui on the frontend. We performed semantic similarity computation with Python scripts using PyTorch.

Challenges we ran into

As we have multiple components both on the front-end (for mobile and web) and the back-end (for feature extraction, summarization and application), providing a smooth communication between the components was tricky!

Accomplishments that we're proud of

In such a short notice, we were able to provide a complete application structure in a modularized way. We had the chance to refresh a diverse set of skills from Rest API, Firebase, Docker applications to front-end technologies.

What we learned

We learned to formulate a real-life NLP problem and convert it to a chatbot application that can be provided to the end user.

What's next for LoGPT

We are aiming to scale our application to live data streams and handle the log information flow in real-time!

Built With

• firebase
• influx
• mui
• nextjs
• node.js
• python
• rest