I applied through other source. The process took 3 months. I interviewed at NVIDIA in Apr 2024
Interview
I recently interviewed with NVIDIA, and I wanted to share my experience, which spanned close to three months. Here are some key points regarding my journey:
Pros:
1. Excellent Interview Panel:
- The interview panel was highly knowledgeable and professional.
- The questions were challenging, particularly around LLMs and the generative AI landscape, which made for a stimulating and engaging discussion.
- The panelists were thorough in their evaluation and made the interview process intellectually rewarding.
2. Positive Initial Feedback:
- I received positive feedback after completing all the interview rounds. - The initial discussions about the offer were encouraging, making me excited about the opportunity at NVIDIA.
Cons:
1. Lengthy and Non-Transparent Offer Process:
- The offer approval process took over a month, during which communication was sparse.
- Despite receiving an initial verbal offer and detailed breakdown of the compensation package, the formal offer was never released on the promised day.
- Eventually, I was informed that the position was put on a hiring freeze, which was extremely disappointing.
2. Impact on Other Opportunities:
- Due to the extended process and optimistic feedback from NVIDIA, I placed other potential job opportunities on hold or declined them, which in hindsight, was detrimental to my job search.
3. Lack of Transparency:
- The HR process lacked transparency and left me in the dark for long periods.
- The abrupt announcement of a hiring freeze after an extended waiting period was particularly frustrating and disheartening.
Interview questions [1]
Question 1
1. In your experience, what are the advantages and disadvantages of fine-tuning a model compared to using the RAG approach? How do you see these methods complementing each other in natural language processing tasks?
2. Can you elaborate on the various techniques available for adapting a foundational LLM to handle domain-specific proprietary data? How do these methods differ in terms of efficacy and practical implementation?