TUM GPT

Inspiration

Getting a response from TUM Informatics Academic Programs Office takes a long wait time.

What it does

It helps to answer queries of students based on the information from TUM website.

How we built it

We leverage LLM to synthesise an answer. To make the answer grounded in sources, we retrieve the actual relevant information, which we provide to LLM to reason and provide the answer.

Challenges we ran into

Wrangling LLMS and end to end pipeline optimizing

Accomplishments that we're proud of

Built an end to end application, which has potential to touch every TUM students need.

What we learned

LLMs are sensitive to prompts, and need craft in prompting, along with its undeterminism makes it hard to catch regex based responses.

What's next for TUM GPT

It has to potential to make TUM, more international student friendly by assisting students with their queries.

Built With

• huggingface
• langchain
• openai
• python