My name is Darshan Makwana, I will be a fourth-year UG in mechanical engineering now. I recently completed my summer internship at Sprinklr and have detailed some of my experiences here
I first heard about Sprinklr through the PT cell via their PPT. Initially, I was not interested in pursuing an internship there but the interview experience changed my mind on what is the actual work that you do there and the amount of ownership you get over your project. The selection process consists of a coding test after resume selection followed by 3 interview rounds
The interview rounds were specific to the candidate’s profile. I had a background in DS/ML/AI so my interviewers were from the AI team of Sprinklr. The best way to prepare is by going over the fundamentals of statistical machine learning and the concepts you used while working on the projects mentioned in your resume
Sprinklr is essentially a SaaS company for CXM. This means that they provide services for managing, growing, and improving customer experiences across social media platforms. Since audio/voice is a crucial way in which humans engage in social media, Sprinklr has a dedicated AI voice team, which I was interning for. My work there was on doing a PoC and training of an audio language model, much like how GPT-4 omni reasons over a shared space of text + audio modalities. They required a unified vanilla model that can reason over both of these modalities while performing a lot of other tasks which will eventually save a lot of inference computing and cost will also increase performance (if it’s scalable)
Since this was a PoC, there were a lot of stagnation points during the internship and many times it was unclear which direction to head on to. The thing that helped us a lot was experimentation and experimentation and experimentation. We experimented with a lot of things from the tokenization, to discrete or continuous encoding of audio for better representation down to the architecture space of the model. This helped us develop an intuition for which methods would work best thus quickly interacting over them. Moreover, I was assigned a mentor and a buddy for the internship, who were extremely friendly and helpful throughout and provided valuable insights whenever necessary. On top of this, we had meetings with our manager to present the project and gather feedback.
I developed birding as a hobby during this time, it started out of nowhere as I watched birds fly across my window during the morning, and later on, I met a few people who were passionate birders. I was able to observe more than 60 birds of various species while capturing more than 30 of them. One thing that I found fascinating was chicks were able to distinguish the calls of their parents from other birds within only a day into the world, this might be an evolutionary advantage they gained over time for survival which is called “imprinting” but some similar observations were found while we were training audio language model wherein the model was able to perfectly recognise the examples that it has only seen once during training, resulting in sharp drops in loss while all we were doing was next token prediction!
Even as an intern you mostly work independently on your project, thus you have to carve out your own action plan/timeline for building things. This also means you have to work long hours in a fast-paced environment but the work is extremely rewarding and growth is immense. But the work you do as an intern never goes to waste and since every project assigned is unique your project is actually pushed to production. If not, then other people learn from your work or try to take the ideas and apply them somewhere else, or refactor your project
If you want to work on hard industry-facing problems with talented and hardworking peers, Sprinklr might be the best way to begin that journey.
