Applied Intuition Software Engineer Interview Experience — AI-Assisted Virtual Onsite Building a Query API over MCAP Driving Logs

This is the same as the earlier thread about this interview.

The format is MCAP, and the content is car driving data, like video, images, location information...

folder structure of the logs:

run_a
  camera
    message_1
    message_2
  location
    message_1
    message_2

MCAP defines a message schema, and the message content is ProtoBuf encoded. Every message has a timestamp, and the API I had to implement needs to query by timestamp, so an index is needed.

My idea was to store the metadata and the raw data separately, but once they are split it is hard to add an index, because every message's timestamp is different, and I also had to implement sync. Maybe I overthought it. I get the feeling the interviewer just wanted everything put straight into a DB.

For AI coding I could use Claude Opus, and it was really powerful. Use Python + SQLite + save every raw message (ProtoBuf decoded) into the DB, and you can have Claude implement all of the required + optional parts directly and generate test scripts. I didn't use the UI, so I don't know whether the time is enough. Have Claude run that "list all runs" and look at the output, and you should understand what is going on. The file they give you is too long and the screen is small, so it was really tiring to read. You can have Claude get started first and then read it slowly. At the end, generate a README.

The presentation is mainly about watching your demo, and they also ask how you interact with the AI.