code2paper

• 

  Demo6

• 

  Demo7

• 

  Demo2

• 

  Demo3

• 

  Demo4

• 

  Bad example

• 

  Query, Context and the actual function

• 

  Reponse from Rag

• 

  Demo

• 

  Demo1

• 

  Query, context and function

• 

  Reponse from Rag

Inspiration

Complex code is hard to read and understand Summaries are easier to understand than going through multiple for loops

What it does

We build a vscode extension that can explain functions in the current file using research paper as context. Use LLMs to analyze the code for you and explain it to you Provide any research paper or technical documentation as context for the generation Context from research paper provides the reasoning for a particular implementation that the code itself doesnt provide often times

How we built it

Convert all the documentation (technical/paper) to embeddings (llama index) and store in vector database (AstraDB) Extension looks for functions in current open file The function that the user needs explanation for is extracted, summarized and the summary is checked in vector database for similarity Results are retrieved and shown.

Challenges we ran into

Tuning RAG chunk size is important Similarity metric matters

Accomplishments that we're proud of

Implemented a vscode extension barebones that summarizes functions for you

What we learned

Tuning RAG chunk size is important Similarity metric matters Top K similar vectors can helpful in giving context but choosing k is difficult

What's next for code2paper

Built With

• astradb
• javascript
• llamaindex