Internship Experience at TCS R&I

Having been introduced to the opportunity to work with TCS R&I for the summer of 2023 at IIT Madras Research Park, I felt like it would be a wonderful experience to have a taste of industrial research. Tata Consultancy Services (TCS) Research and Innovation (R&I) is the research arm of Tata Consultancy Services, one of the world's largest and leading IT services, consulting, and business solutions organizations. TCS R&I plays a crucial role in driving technological advancements, innovation, and thought leadership within the company and across the industry.

I had applied through campus, and the selection process involved the shortlisting of resumes followed by an interview. The domain I wanted to work on was physical sciences and the desired location was filled as Chennai. There were some delays that existed in the scheduling of interviews, however, the call was received around a week later than the scheduled dates. The interview lasted around half an hour. The questions mostly dealt with my resume and projects that I had done/been doing then/now. Some questions were along the lines of courses completed and my knowledge of coding, ML, and NNs. It was quite obvious then that the problem statement would be some kind of integration of my core field (mechanical engineering) and other domains mentioned above.

I received the internship offer letter on May 3rd indicating that the interviewer was happy with the performance and was introduced to the problem statement ‘Utility of Physics Informed Neural Networks to model fluid flow in buildings’. The internship was to officially commence in May-15-2023 and I had to report at IIT Madras Research Park for the same during the summer.

The first day basically involved the completion of certain formalities and getting acquainted with my mentor Dr. Sri Narayana Nagarathinam over tea. My knowledge in Fluid Mechanics, Heat & Mass Transfer was tested along with my understanding in ML, and NNs. The problem statement, scope, and expectations were discussed at length.

Coming to the problem statement, if one were to ask me what I did without getting too technical, an attempt was made to model the fluid flow and evolution over time for different geometries, and initial and boundary conditions. The idea is to solve PDEs/ ODEs using NNs, by incorporating the PDEs & boundary conditions & initial conditions into the loss function and trying to minimize it, hence the name ‘Physics informed Neural Networks’ (PINNs). The specific set of PDEs dealt with were obviously the Navier-Stokes Equations. My first task was to implement the classic pipe flow problem for both steady and unsteady cases of the system. My second task was to implement the well-known benchmark problem in CFD, the ‘Lid Driven Cavity’. My final task was to implement PINNs for certain room geometries each having velocity boundary conditions, temperature boundary conditions, and initial conditions.

While a workplace was in the process of being allotted to me, Codes were written on my laptop for the initial tasks. However, even after the allotment of a workplace with a desktop, I soon realized the allotted place did not run the codes fast enough to actually make a 1-hour trip to the research park worth it. Working on my laptop was a better option given the time constraints and the durations the codes were taking to run. Sometimes I had to run some codes in my laptop that lasted 2 or more days only to get not-so-satisfactory results. This meant it would have taken longer had I run those at the allotted place. Consequently, I discussed with my mentor and mutually agreed on having weekly meetings with updates on the work done. Meetings with my mentor turned out to be fruitful and enriching as he always had something to add to what was done or suggested new ideas. They were more like brainstorming sessions and along with my mentor's patience and willingness to entertain suggestions/ideas from my side, these meetings proved to be extremely informative.

As my official internship end date was nearing I chose to extend it by a week as the problems dealt with became more complicated leading to more training times causing the durations for hyperparameter tuning to reduce. Simultaneously, I spent my time preparing the internship report to be submitted to my mentor. The act of extending the duration not only helped me achieve satisfactory results but also allowed me to spend more of my time on the report which was going to be the one conveying the progress made in a detailed and formal manner to my mentor.

Looking back at how my summer went, I would say it was an enriching experience having worked on a topic that has only recently been explored, still in the realm of fundamental research.