The process occurred through on-campus placements. It consisted of four rounds: an Online Assessment (OA), two Technical rounds, followed by an HR round.

• OA: Included four Machine Learning/Python Multiple Choice Questions and one end-to-end implementation of RAG using LangChain.

• Technical Round 1: For my experience, given my multiple ML internships and projects, this round focused entirely on my resume and Deep Learning basics to intermediate concepts. Other candidates were presented with a medium Data Structures and Algorithms (DSA) question.

• Technical Round 2: Due to a strong performance in Round 1, my second technical round focused entirely on a High-Level Design (HLD) problem, specifically improving an existing Agentic pipeline. Other candidates were asked resume-based questions.

• HR Round: This was a friendly discussion, covering work aspects and location preferences.

What is the difference between BERT-based text embeddings and Word2Vec?

In the case where a tokenizer breaks down each word into multiple (excessive) tokens, what is the effect on downstream tasks such as Sentiment Analysis and Named Entity Recognition (NER)?

What is the difference between using positional embeddings from a patch-CNN (trainable) versus the non-parametric ones used in the original "Attention Is All You Need" paper?

Questions around optimizing Voice-to-Text and Text-to-Speech models and the effects of model quantization on their inference quality.