Top Tech Transition ₹9 Masterclass

Real Interview Experiences

Learn what to expect, straight from candidates who've been through it at top tech companies.

1346 interviews325 companies432 offers
Loading experiences…

Browse by company

Browse by role

← Back to all experiences

Cisco Agentic AI Developer (Grade 8) Interview Experience

Cisco · Mid level · AI Engineer

I hadn't really practiced doing an agentic system design round in an online drawing tool, so that caught me unaware...
Result—
Timespan—
DifficultyDifficult
Rounds2

Interview process

I interviewed for an Agentic AI Developer role at Cisco, grade 8, and it was only two rounds. The first round was Python-heavy: two basic DSA problems, one around binary trees and one simple array question, plus a role-specific problem where I had to return the top three relevant documents from a set of 10 to 20 prompts using my own RAG-style chunking approach. The second round was much more domain-specific and centered on architecting a system where a user queries an LLM, the system finds relevant docs, and then returns a response. They also went into storage internals and my past work, so it definitely made me think a lot. What felt different was how focused the whole thing was on agentic AI instead of just a standard software interview.

Interview rounds · 2

  1. 1

    Technical round

    CodingData Structures & AlgorithmsArtificial Intelligence

    The first round was basically a Python coding round. I got two basic DSA questions, one on binary trees and one simple array problem, and then a more role-specific word problem where I had to build my own RAG-style chunking logic.

    1. Q1. Solve a binary tree problem in Python.
      How they answered

      I solved a binary tree DSA problem in Python. It felt like one of the two basic coding questions in the round, and they were pretty clearly looking for clean Python problem solving first before getting into the more agentic AI-specific stuff.

    2. Q2. Solve a simple array problem in Python.
      How they answered

      I also got a simple array-type problem and solved that in Python. That one was more straightforward and felt like a basic screen to check that I could code comfortably before moving to the role-specific question.

    3. Q3. Given 10 to 20 different prompts as input, return the top three specific documents in order of relevance.
      How they answered

      For the word problem, I had to implement my own RAG-based chunking system in Python. The input was around 10 to 20 different prompts, and the output had to be the top three specific documents ranked by relevance. That was the most role-specific part of the first round.

  2. 2

    Final / onsite round

    System DesignArtificial IntelligenceMachine LearningTechnicalProject Discussion

    The second round was the real domain round. I had to draw an agentic RAG system live, then go deeper on storage internals, and they also drilled into my past work and agentic frameworks.

    1. Q1. Architect a system where a user queries an LLM, the system finds the relevant documents, and then returns the response.
      How they answered

      I had to build an agentic RAG-based system live using an online drawing tool. I walked through how a user query would hit the LLM, how the system would retrieve the relevant documents, and how the response would come back. Then they pushed deeper into memory, RAG flow, scaling the database layer, and security around the agent. That part was different from the high-level design rounds I had practiced because they wanted more agent-specific detail.

      Follow-up questions
      • How would you handle memory in the agentic system?
      • How would you handle RAG, database scaling, and security for the agent?
    2. Q2. Walk through the storage layer of your RAG system: how documents are indexed and retrieved, such as with an inverted index or vector store.
      How they answered

      They asked me about the internals of storage and specifically got into inverted database concepts. This was less about broad architecture and more about whether I actually understood how retrieval-related storage works under the hood.

    3. Q3. Define agentic.
      How they answered

      They asked about my past projects and then drilled down into the agentic part of them. I had to explain what I mean by agentic, and they asked what frameworks I would use, including things like LangGraph and orchestrator-style setups. It felt like they were checking whether I could go beyond the buzzword and talk concretely about how these systems are actually built.

      Follow-up questions
      • Tell me about agentic projects you've worked on.
      • What frameworks would you use? LangGraph or an orchestrator?
      • Why would you choose those frameworks in an agentic setup?

Tips from the candidate

I hadn't really practiced agentic system design rounds in an online drawing tool, and that showed me what I should have prepared better. I'd focus less on only generic high-level design and more on how an actual agent system works, like memory, RAG, scaling the database side, and security around the agent. I'd also make sure I can explain what agentic really means and talk through frameworks like LangGraph and orchestrator-style setups without staying at a buzzword level.

Company culture

They weren't doing a generic interview loop with me. It felt like they wanted practical agentic AI knowledge and would keep drilling until they could tell whether I actually understood the system, especially retrieval and storage details. Even the coding round was tailored toward the role, so they seem to be screening for people who can build this stuff, not just solve standard DSA questions.

Details

CompanyCisco
RoleAI Engineer
LevelMid level
LocationUnited States
InterviewedJun 2026
Questions asked6