Skan AI

1. 1

   Technical interview

   Round 1: ML fundamentals, project deep dive, paper reading
   60 minhard

   project design choices · loss functions · evaluation protocol · attention mechanism · RNN and LSTM · vanishing gradients · skip connections · reading a new research paper

   Interviewer was a data scientist. About 35 minutes of fundamentals, then 15 minutes silent paper reading, 5-7 minutes discussion. Candidate was asked about parameter counts and failure modes of own project.

2. 2

   Technical interview

   Round 2: project grilling and open-ended system design
   hard

   cross attention · multi-head attention · training-free methods · custom loss terms · LLM-driven web automation agent design · failure recovery

   Interviewer was a data science director. Open-ended company problem with no fixed answer; goal is visible thinking.

3. 3

   HR interview

   Round 3: culture fit

   career goals · why this company · startup vs legacy company · logistics and compensation

   By the end of this round the offer was clearly coming; logistics and compensation were discussed.

Advice from candidates

98 pieces of advice, in their own words. Names removed.

• The platform is very secure and strict. It will detect any head movement or noise. You have to share your screen as well. So, choose a place that is very silent and don’t let anything distract you.

  Test · 2023

• Be honest and well-prepared about your resume. I gave real world examples to explain my answers.

  interview · 2023

• The DSA questions in the coding round were easy, but the ML questions were pretty hard. But, I think that people who solved the DSA questions were shortlisted for the interview.

  Test · 2023

• For OOPS do these 40 questions. All OOPS questions were from these only.

  preparation · 2023

Loading the other 94…

Question patterns reported

The kind of problem, not the problem itself. Our practice questions are written to match these patterns.

• Codingexpanding window on 0/1 array with direction ambiguity
• Codingmax connected value in grid via BFS or DFS
• Codingimplement heapify and pop in Python
• Technicaltwo vague ML application problems needing clarification
• Onlinearray scan with two-sided look-ahead collection rule
• Onlineimplement max heap
• TechnicalBFS in code, then iterative with a stack
• Technicalclassify daily, monthly and chance activities from a time log
• Onlineone basic BFS, two medium DP questions
• Technicalconditional probability paradox puzzle
• Technicalkth largest in unsorted array, quickselect
• Technicaldetect loop in linked list
• Technicalgroup screenshot text boxes by proximity to extract fields
• Technicallongest substring without repeats, sliding window
• Technicalcount islands in grid
• Technicalposterior probability of fault after positive scan
• Technicalexpected flips to see a two-flip pattern, Markov chain
• Technicaledit distance DP
• Technicalcount solutions of three-variable sum equation
• Technicalfind most frequent action sequences of length 3-6 across days
• Technicalgroup unknown multi-colour objects into bins as they arrive, colours hashed
• Technicalrecover personal details from a screen recording of form filling
• Technicalwhy skip connections help CNNs but not plain RNNs
• Technicalread unfamiliar paper in 15 minutes, then discuss