Ema Interview Experience

You are given a directed graph having ‘N’ nodes. A matrix ‘EDGES’ of size M x 2 is given which represents the ‘M’ edges such that there is an edge directed from node EDGES[i][0] to node EDGES[i][1]. Find whether the graph contains a cycle or not, return true if a cycle is present in the given directed graph else return false. For Example : In the following directed graph has a cycle i.e. B->C->E->D->B. Note : 1. The cycle must contain at least two nodes. 2. It is guaranteed that the given graph has no self-loops in the graph. 3. The graph may or may not be connected. 4. Nodes are numbered from 1 to N. 5. Your solution will run on multiple test cases. If you are using global variables make sure to clear them. Input Format : The first line of input contains an integer 'T' representing the number of the test case. Then the test cases are as follows. The first line of each test case argument given is an integer ‘N’ representing the number of nodes in the graph. The second line of each test case contains a given integer ‘M’ representing the number of edges. The next ‘M’ lines in each test case contain a matrix ‘EDGES’ of size M x 2 which represents the ‘M’ edges such that there is an edge directed from node EDGES[i][0] to node EDGES[i][1]. Output Format : For each test case, print true if a cycle is present in the given directed graph else print false. Note : You do not need to print anything; It has already been taken care of. Constraints : 1 ≤ T ≤ 5 2 <= N <= 100 1 <= M <= min(100,N(N-1)/2) 1 <= EDGES[i][0], EDGES[i][1] <= N Where ‘T’ is the number of test cases. Time Limit: 1 sec

Machine Learning

In an agentic AI system like Ema’s that handles long-running conversations and performs multiple tool-calling steps, explain why KV (Key-Value) caching is important for efficient inference. Discuss how it helps with maintaining context across interactions, reducing recomputation, and improving response speed. Also describe the trade-offs involved, particularly in terms of increased memory usage, scalability challenges, and latency optimization. Standard self-attention has a memory complexity of O(n2)O(n^2)O(n2) with respect to sequence length. Explain how FlashAttention optimizes the attention mechanism to reduce memory overhead and improve efficiency. Why is this considered a major breakthrough for deploying agentic AI systems that require long context windows? In Reinforcement Learning, explain the key components of a Markov Decision Process (MDP) and their roles in defining an environment for decision-making.

I believe I was rejected because I was unable to answer some of the ML questions they asked. Also, the interview had less emphasis on DSA compared to typical SDE companies, and they seemed to expect candidates to have an ML background.

Key Preparation Tips

Prepare DSA thoroughly.

Practice aptitude and puzzle-based questions.

Have at least two good projects on your resume.

Be prepared to explain your projects clearly.

Practice coding problems involving Arrays, Binary Search, DP and Recursion.

Prepare common HR questions such as Tell me about yourself and Who is your role model?