Ho presentato la mia candidatura online. La procedura ha richiesto una settimana. Ho sostenuto un colloquio presso C3 AI nel mese di set 2023
Colloquio
From the outside, the C3 AI hiring team appears to be disorganized and poorly managed. Here is my story.
– Submitted resume online.
– Received link to 75-minute HackerRank online assessment after a day. Online assessment contained 9 questions, including 1 coding exercise and 8 multiple-choice questions testing knowledge of statistics, probability and machine learning.
– 1 day after completing online assessment: Received invite to schedule 30-minute video interview.
– 1 day before scheduled interview: Received generic email from a no-reply server stating interview cancellation due to shift in company hiring priorities.
Domande di colloquio [7]
Domanda 1
An equipment supplier wants to bundle items that are ordered frequently and repeatedly together. Assume the orders from a customer are placed as 'AABCBAABD', the supplier wants to know the longest continuous ordering sequence such that each item has been ordered at least k times.
For example, if k = 2 and orders = ‘AABCBAABD’, the longest continuous ordering sequence will be ‘BAAB’. In case there are multiple sequences of the same length that all satisfy the requirement, return the result as a sorted list of strings in alphabetical ascending order. If there is no sequence such that each item was ordered at least k time, return an empty array [].
Which of the following is a disadvantage of using convolutional neural networks for tabular time series data instead of a recurrent neural network?
a. Convolutional neural networks are more likely to run into vanishing/exploding gradient problems.
b. Convolutional neural networks have less flexibility in the time dimension.
c. Convolutional neural networks are more computationally expensive than recurrent neural networks.
d. Convolutional neural networks would perform convolutions along the feature dimensions which often do not make sense.
When building an ID3 decision tree, X decides to encode information using a die sample space of {1, 2, 3, 4, 5, 6} instead of a coin sample space {0, 1}. What is the value of ‘b’ X uses in the equation for entropy?
a. 2
b. 3
c. 6
d. 10
e. e
Maria has 100 marbles, 50 of which are blue and the remaining 50 are green. Which of the following statements is true of this marble set’s color parameter?
a. Impurity index of Entropy = Impurity index of Gini coefficient = 0
b. Impurity index of Entropy = Impurity index of Gini coefficient = 1
c. Impurity index of Entropy < Impurity index of Gini coefficient
d. Impurity index of Entropy > Impurity index of Gini coefficient
You are building a nuclear missile detection model which is a binary classification on radar images. What metric is especially critical to consider?
a. Recall
b. Specificity
c. F1 score
d. Accuracy
On average, about 20% of the local population take a vaccine before the flu season. A person who takes the vaccine has a 1% chance of getting the flu. Those who did not take the vaccine may experience an elevated probability of 5% in getting the flu. If a patient in a hospital is suffering from the flu, what is the probability that the patient took the vaccine?
Ho presentato la mia candidatura online. Ho sostenuto un colloquio presso C3 AI (Singapore)
Colloquio
Hackerrank --> three tech interviews (proceed to the next one if you pass the current one) each round is 1 hour long --> hiring manager interview (1 hour)--> VP interview.
Domande di colloquio [1]
Domanda 1
tech interviews: 1) (1 hour) traditional ML based case study, 2) (1 hour) ML concept deep dive, and 3) (1 hour) coding (leet-code medium)
Ho sostenuto un colloquio presso C3 AI (New York, NY)
Colloquio
Resume screening -> technical assessment -> 4 rounds of interviews:
- personal projects, simple questions not there to trick you
- situational questions: "what would you do if..."
- machine learning: starts from the very basics (stats and probabilities) to more up to date models
- coding: medium leet code
Ho presentato la mia candidatura online. La procedura ha richiesto 3 settimane. Ho sostenuto un colloquio presso C3 AI (Londra, Inghilterra) nel mese di ott 2025
Colloquio
I applied directly after seeing a job advert on LinkedIn. There are MCQ and coding assessment on Hackerank, followed by a screening interview. It all went well and got invited to the technical day.
To prepare for the technical interview, I went through all materials and questions shared by others on this website and once I was half way, I noticed that the questions tend to be similar, except the pairwise coding. I recommend you go through questions here to be better prepared for the technical day.
The interview was generally okay and the team was nice. Started off with Case Study (30 mins); followed by ML questions (30 mins); and finally coding (1 hour). There is barely time in-between to switch so expect to transition very quickly. For the case study, think out loud it helped me to figure the actual problem, as they only share the problem and you figure the rest out.
The coding was fair, I had done a couple of Leetcode but they started off with Linear regression etc, kinda caught me off guard and wasted 35 mins on it. Though the program ran, the interviewer said there isn't enough time to complete second question, and we shared our coding experiences and clarity on a few questions. I am pretty confident in stats and ML knowledge but the issue could have been coding; so make sure you are up to speed with anything that can be thrown at you.
Two days later I received a rejection email. No reason after having spend so much time is a bit disrespectful but we move on.
Domande di colloquio [1]
Domanda 1
Case study: Waste reduction in chain stores. They simply stated that and I described it as a demand forecasting problem that can be solved with Linear Regression. Besides clarification questions, It was fine and they took it.
MLQ
1. Difference between Supervised and Unsupervised Learning, and give examples
2. Difference between bagging and boosting;
3. Bias and variance, and explain in the context of Bagging/boosting
4. Performance metrics; what does AUC mean, interpret AUC of 50%
5. Gradient descent
6. Overfitting and Underfitting and how to overcome them in Decision Trees
Coding: Implement linear regression, numpy, and plotting importance scores
