Concentric.AI | Bad Experience with Architect Round
Summary
I went through four technical rounds with Concentric AI in Bangalore, covering DSA, merging K sorted lists, a payout‑rule implementation, and a metadata system design, but the final architect round was vague and left me frustrated.
Full Experience
Concentric AI Bangalore had total 4 Technical Rounds. All 4 rounds were online
1st Round (1 Hr): Basic DSA question (I don't remember exact question but it was Leetcode medium)
2nd Round (60 mins) merge K LinkedList : https://leetcode.com/problems/merge-k-sorted-lists/
2nd Question I don't remember
3rd Round (60 mins) (Online): Discussion on previous projects, and asked me to implement below in my IDE Payout Rule - engine Requirments: Block or accept payment based on rules eg. if vendor amout for last 7 days is more than 10Lac then reject Req: {vendor_id, amount, timestamp} Response: {status: ACCEPTED/REJECTED}
I was able to implement it quickly.
Interveiw 4 (60 mins) (online) It was taken by an Architect (so called), first time he cancelled after making me wait for 15 mins in interview and then rescheduled.
When interview actually happened, he first of all didn't show his face, he first talked on one of my project where he was interrupting me multiple times (asking questions like why Mongo which I had already explained, I guess he was disctracted when I was explaining).
Then gave very vague System design question, after asking a lot of follow up questions I reached to below problem statement, I explained each component but he was not at all interested in listening tradeoffs or have a discussion, and communication skills were so bad, I was asking a question he is saying hmmm, what the hell, who makes such people Architects
I was working through the problem then he abruptly says we need to close, I asked multiple questions to him regarding company but he gave answers this all we are doing, no depth, just throwing vague names.
I am really frustrated, took so much time giving 4 rounds and I am sure such companies are just wasting time of candidates, they don't want to hire anyone.
Design a Metadata System -> user has different types of files (like video) already uploaded -> process the video file -> create metadata from this video (like multiple resolutions, ,size, transacription, who has access) -> Analytics on this data (like who created most no of videos in last one week)
Non Functional:
- Highly available
- Latency should be as minimal as possible (precompute vs computing on runtime)
Scale Requirements: -> No of files: few TB's of data -> 50M files existing -> 100k new files coming everyday -> 600 rps to metadata service -> 2-3 hrs of dealay
DB: Read heavy + very large scale
Interview Questions (3)
Merge K Sorted Lists
Given an array of k linked‑list heads, merge them into one sorted linked list and return its head.
Payout Rule Engine
Implement a payout rule engine that decides to ACCEPT or REJECT a payment based on vendor history. Example rule: if the total amount paid by a vendor in the last 7 days exceeds 10 Lakh, reject the payment. Input: {vendor_id, amount, timestamp}. Output: {status: ACCEPTED/REJECTED}.
Design a Metadata System
Design a highly available metadata system for uploaded video files. Requirements include processing videos to generate metadata (multiple resolutions, size, transcription, access control), providing analytics (e.g., who created the most videos in the last week), handling read‑heavy large‑scale data (≈50 M existing files, a few TB total, 100 k new files per day, 600 RPS to the metadata service, latency minimised, 2‑3 hour delay tolerance). Include discussion of storage, indexing, scalability, and non‑functional constraints.