I applied for the Cresta Forward Deployed Engineer (FDE) role, on the AI Agent side.
I just finished the OA, and it is very different from the usual pure-algorithm OA. Overall it leans toward practical coding, understanding requirements, and real engineering ability. Sharing the question types for reference.
The OA has 5 parts in total, each one timed separately, with these limits:
10 / 30 / 20 / 25 / 18 minutes
The timing is pretty tight overall, especially for the later questions. If you start out with an overcomplicated direction, you will barely have time to start over.
Q1: AI project recording — 10 min
You have to record yourself talking about an AI project you worked on before, covering what the project was and which part you were responsible for.
It felt more like a short project / behavioral screening, not coding.
I talked about an AI/data-related side project I had done before.
Q2: File system / path — 30 min
You are given a folder / directory, and you need to find a certain file according to the requirements and return its absolute path.
It is more like practical Python programming than an algorithm question. If you are comfortable with os / pathlib and the like, it should go smoothly.
Q3: 2D grid simulation — 20 min
You get a 2D array where each value is the number of leaves, plus a sequence of wind directions.
You simulate how things change according to the wind direction, and in the end calculate how many leaves are left.
It is essentially a simulation. The algorithm itself is not especially hard, but with only 20 minutes you need to figure out quickly how the state updates.
Q4: Message decoding / string matching — 25 min
You are given a set of known words and an encoded message, and you need to decode the message back using those known words.
The setup is roughly that an encoded word is made by rearranging the letters of the original word, so at first I used each word's 26-letter occurrence/frequency counts as the encoding key.
The annoying part: the problem description made me understand that there would be no anagram ambiguity, but the visible tests actually did contain anagram cases.
In the end I used the letter-frequency encoding plus the first letter of the encoded token together as the mapping key. That passes the visible tests, but I am not sure about the hidden tests.
Q5: Prefix / suffix overlap — 18 min
You are given two words/strings and asked for the maximum length by which the end of one string can overlap with the start of the other.
Essentially you can start from the maximum length and check whether a[-k:] == b[:k].
Discussion
Loading comments…