CV

Aim was to research with sequential models such as LSTMs in order to predict outputs using batches. This involved batching input data, as well as comparing the prediction performance as well as usability (training and inference time). Setup an experiment flow running the training on Modal and logging results to WandB.

There is a part where a given state space needs to be explored for possible solutions. In order to achieve greatest coverage, random sampling is currently used, which however becomes weaker once conditions are applied on the state space variables. Aim was to benchmark and PoC additional sampling techniques to improve sample generation.

Building on the work done by the ISE Lab in 2019 called Spindle, I worked on improving the performance (accuracy and mean F1) of the model by experimenting with different State of the Art vision models such as Vision Transformers, data augmentation techniques and sequential processing layers. Read more about it here!

Worked as part of the team to conceptualize and start building Modyn , a platform and benchmark tool aimed specifically to handle the problem of retraining models during continual learning.

Extended the previous study to further classify Cancers according to the CIMP subtype. This was done by applying the existing methodology on new datasets, as well as extension of the analysis to find further genetic pathways of interest.

As part of the ETH Analytics Club, worked the NGO BASE to analyze data generated from their app and provide reports that could be used by the NGO to monitor the performance. Additionally worked with them for a month to create a service to help provide this data directly to the application via REST Apis.

Selected to be part of the cloud computing TA team in order to help conduct exercise sessions, help redesign and manage the practical project as well as answer any student queries either in the sessions or on the online forum.

All engineers took turns answering technical requests and bugs raised by clients. Involved looking through code and logs, working with different teams, fixing issues (short and long term) as well as replying / working with clients to handle the issue.

Helped build a health dashboard page for client sales integrations to provide clients with information relating to the health of their integration along with action items they can take to improve the health and avoid future complications.

Helped build a new integration for TradeGecko allowing clients to connect their eBay stores. We worked directly with eBay APIs as well as built an Open Source gem to help others connect to eBay.

After TradeGecko was acquired by Intuit, lead the work to redesign and update the Onboarding page to work for new clients coming directly from QuickBooks.

Helped perform a more in depth analysis of US Tax calculations and their flows from Shopify into TradeGecko and into QuickBooks.

Enhanced and added new client reports that would be sent out to clients daily. Edited SQL Stored Procedures as well as XML templates to correctly populate reports. Additionally helped build support tools to allow non technical teams to update and manage client reports on their own.

Setup a new Redis cluster in order to add redundancy and improve performance of the GS Investment Research website.

Setup new calculations in SLANG in order to compute additional metrics based on Analyst inputs, and also helped build the flow to ensure their timely update on the Investment Research Website.

As part of a community outreach step, we started an initiative in the Singapore office to organize simple coding workshops for secondary school kids using Raspberry Pis. The aim, to help encourage the next generation of coders, and also to encourage more females to be interested in coding.