Interview Exp | Emergent Labs | Rejected | ~3Yrs Exp
Summary
I interviewed online with Emergent Labs for a machine coding round, was asked a design/implementation question about routing LLM calls, and was rejected after the first round.
Full Experience
I recently gave an online interview for Emergent Labs and got rejected in the first round itself. It was a maching coding round . The syllabus mentioned is quite exhaustive and they do really ask from the same set . Syllabus: https://interview-prep-101.emergent.host/#machine-coding
They asked me a story based question on Circuit Breaker.
Question: You own a LLM orchestrator system where all calls go to Claude. If you start getting error, after 5% , divert rest 95% to openAI . If the error calls go above 90%, respond back with error while 5% requests go to Claude and openAPI each.
You need to write production level working code , explaining entities and design patterns with the help of AI.
I told the interviewer my understanding of the question, the entites and design patterns (State, Strategy).
**They asked how will I decide how to route the request to a particular LLM? **
After 5-10 mins of discussion , going back to & fro, I told we will hash the requestID of each call and mod by 100 . Basically consistent hashing , he agreed but maybe I spent a lot of time here.
At the end, he told me to generate the code using AI. They particularly note your prompting skills and see how accurate code you can generate from it. Obviously on the usage of AI, you need to drive it.
At the end, after running code, there were 1-2 bugs in the code which resulted in few inaccuracies. Attaching arch of the code files:
Interview Questions (1)
LLM Orchestrator Routing with Circuit Breaker
You own a LLM orchestrator system where all calls initially go to Claude. If errors start occurring, after 5% error rate you should divert the remaining 95% of traffic to OpenAI. If the error calls exceed a 90% threshold, the system should respond with an error while still routing 5% of requests to Claude and 5% to OpenAI. The task is to write production‑level code that implements this behavior, explains the involved entities, and applies appropriate design patterns (e.g., State, Strategy). Additionally, you were asked how to decide routing for each request.