Candidates for Citi AI Engineer roles describe a Karat or Codility screen followed by a technical conversation on GenAI architecture or past AI projects. Here is what they report and how to prepare.
$138K
Avg. Base Comp
$165K
Avg. Total Comp
Not reported
Typical Rounds
Not reported
Process Length
Citi AI Engineer interviews, as three candidates describe them, pair a coding screen with a conversation about real AI work. Two senior GenAI candidates report a Karat coding round. One described a 1-hour session: about 10 minutes of basic Python, then roughly 40 minutes on two coding problems. That candidate solved the first and found the second much harder. The reported problem was a graph question. Each location has one outgoing road leading to a campground, two cars start at different points, and you must work out which people each car picks up, including ties on arrival time.
The second senior candidate, who applied through Workday, did the Karat round and then met an Architect Lead. That conversation centered on how to design a production RAG system: document ingestion, chunking, embeddings, retrieval, LLM generation, hallucination handling, evaluation, security and monitoring. They also discussed scalability and design tradeoffs. This candidate felt confident on architecture but found the coding harder.
A third candidate, in an AI/ML specialist role, reported a different path: application screening, a Codility assessment, then a 60-minute behavioral and technical interview. The technical portion focused on their own AI projects, with no algorithm puzzles. Questions covered motivation for joining Citi, teamwork and integrity, and how they thought about feasibility and implementation. They received an offer. The two senior candidates were still waiting to hear back.
The practical takeaway is to prepare on two fronts. Practice graph and traversal problems under time pressure, and be ready to explain a RAG pipeline end to end. Also prepare to discuss your own AI projects, including metrics, tradeoffs and lessons learned. Formats may vary by level and team.
Synthesized from 3 candidate reports by our editorial team.
Had an interview recently?
Share your experience. Unlock the full guide.
Real interview reports from people who went through the Citi process.
Applied through Workday, completed a Karat coding round, then interviewed with an Architect Lead. The process was structured and technical. I felt confident discussing GenAI architecture, but coding was challenging.
Questions asked: The Architect Lead asked me to explain how I would design a production RAG system, including document ingestion, chunking, embeddings, retrieval, LLM generation, hallucination handling, evaluation, security, and monitoring. We also discussed system design tradeoffs and production scalability.
Share your own interview experience to unlock all reports, or subscribe for full access.
Sourced from candidate reports and verified by our team.
Topics based on recent interview experiences.
Featured question at Citi
Write a query to show the number of users, transactions, and total order amount per month in 2020
| Question | |
|---|---|
| Size of Joins | |
| Attribution Rules | |
| Jars and Coins | |
| Hurdles In Data Projects | |
| Above Average Product Prices | |
| Real-Time Transaction Streaming | |
| International e-Commerce Warehouse | |
| Concurrent LLM Serving | |
| Dijkstra implementation | |
| Client Solution Pushback | |
| Logistic Regression from Scratch | |
| Replace Words with Stems | |
| String Palindromes | |
| Loan Model | |
| Seller Type Modeling | |
| Index Fund Return | |
| Text Editor With OOP | |
| Unified Event Pipeline | |
| Data Cleaning Experiences | |
| Why Do You Want to Work With Us | |
| Your Strengths and Weaknesses | |
| Concentric Circles | |
| Singly Linked List | |
| Student Tests | |
| Measuring Text Difficulty | |
| Optimizing Threshold Adjustment in Default Risk Models | |
| 2nd Highest Salary | |
| Empty Neighborhoods | |
| Rolling Bank Transactions |
Synthesized from candidate reports. Individual experiences may vary.
Candidates report applying online, with one mentioning Workday. One candidate describes an initial application screening before any assessment. Timing between steps was not reported, so expect to wait for recruiter contact.
Two senior GenAI candidates report a Karat coding round. One described 1 hour: about 10 minutes of basic Python, then about 40 minutes on two coding questions, including a directed-graph pickup problem. A third candidate reports a Codility assessment of coding fundamentals instead.
One senior candidate reports meeting an Architect Lead after Karat. The discussion covered designing a production RAG system, from ingestion and chunking to retrieval, generation, hallucination handling, evaluation, security and monitoring, plus scalability tradeoffs.
One AI/ML candidate reports a 60-minute behavioral and technical interview. It included tell-me-about-yourself and why-Citi questions, then a technical discussion of their own AI projects, covering metrics, tradeoffs, feasibility and implementation. That candidate received an offer.
