The interview started with some self-assessment questions. I was asked to rate myself in Python and Agentic AI, and briefly explain my experience in both.

The interviewer then went deep into my AI projects and asked implementation-level questions, including:

1. How RAG works end-to-end
2. How chunking and embeddings work and how embeddings are generated
3. How vector databases and similarity search work internally
4. How cosine similarity, KNN and ANN work
5. How HNSW works for efficient vector search
6. Sub-agents vs tools and how an agent decides which tool to use
7. How MCP works, including its architecture and implementation

After that, I was given a Hard-level DSA problem:

LeetCode 761 — Special Binary String

The question involved identifying special binary substrings and rearranging consecutive special substrings to obtain the lexicographically largest string.

Example:

Input: "11011000" Output: "11100100"