Sharing my ML engineer interview experience with @ringg_ai
(Funded voice AI company)
Base 65LPA + equity + bonus
It was an one hour interview headed by the chief AI officer and CTO in their office.
And honestly the bar couldn’t be higher.
Started directly with my internship experience where I led the Speech pipeline at a commerce Robotics in tokyo.
We had a deep dive into how I reduced the p95 latency to ~800ms and then we discussed a bit on using spacy to reduce the WER (increase latency)
We had a deep dive into how I reduced the p95 latency to ~800ms and then we discussed a bit on using spacy to reduce the WER (increase latency)
Then we moved onto my ongoing project of training directly speech to speech transformer instead of traditional (speech -> text -> LLM -> text -> speech)
That time i was fine tuning qwen 3 decoder only model, but the loss wasn’t decreasing and he guided me some tricks to try.
That time i was fine tuning qwen 3 decoder only model, but the loss wasn’t decreasing and he guided me some tricks to try.
Then we deep dived into my current role of data scientist and man he was smart.
We discussed in detail how to model credit risk and also the nitty gritty details of feature selection using PCA instead of traditional recursive elimination, isolation forest and so much more.
We discussed in detail how to model credit risk and also the nitty gritty details of feature selection using PCA instead of traditional recursive elimination, isolation forest and so much more.
After this we went through the networking concepts as why we need queues when we can directly use web sockets for voice bots.
And we also covered a bit of GPU parallelism, more specifically FSDP2 and I coded the architecture of it.
And we also covered a bit of GPU parallelism, more specifically FSDP2 and I coded the architecture of it.
Result : No hire
But it was one of the many turning points of my career.
This was in Jan’26.
But it was one of the many turning points of my career.
This was in Jan’26.
Generated by Thread Navigator
Press ⌘ + S to quick-export