May 2023 – Oct 2024 · 1 yr 5 mos · Delft, South Holland, Netherlands · Hybrid

• As a part of a team of freelancers, I work on applying artificial intelligence to various problems in the urban sector. Most notably, we built a retrieval augmented generation (RAG) based chatbot for multiple municipalities in the Netherlands to handle user queries easily and more efficiently, increasing response efficiency (number of requests handled successfully per week) by 40%. I worked with Langchain, Llama-3 and GPT 4 models and APIs, FAISS and ChromaDB vector databases. I built the frontend to be simple and intuitive using Streamlit UI and an alternate prototype using Chainlit. The deployment was on GCP with Docker. Apart from development, deployment and testing, I also enjoyed communicating and presenting the results to the clients.

Aug 2021 – Jul 2022 · 11 mos · Bengaluru, Karnataka, India · Hybrid

• As a part of the Advanced Queueing team, I was responsible for the development and maintenance of the AQ systems. This was also my first full time experience at a big organisation, where I improved my ability to work in a team.

Feb 2021 – Jul 2021 · 5 mos · Bengaluru, Karnataka, India · Remote

• As a part of my bachelor thesis, I worked on a tool to identify galactic cirrus clouds. I built an auxiliary tool to convert ICRS coordinates to galactic coordinates and vice versa, and used YOLO v2 algorithm for detection. I built the train and test dataset using NASA's Wide-field Infrared Survey Explorer All-Sky Dataset. I performed the majority of the operations such as handling FITS files using the Astropy library. The final model was able to detect 96% of the galactic cirrus clouds, with no false positives.

Apr 2020 – Jul 2020 · 3 mos · Bengaluru, Karnataka, India · Remote

• I worked on making the selection phase of a genetic algorithm faster using off-policy evaluation techniques. I attempted to estimate the fitness of a section of individuals in a population using GAE and KL-divergence, while the remainder were evaluated on-policy. There was a ~10% improvement in the evaluation time for the selection phase, but it was not consistent across multiple trials and usually was achieved at the cost of accuracy.

Jun 2019 – Jul 2019 · 1 mo · Kolkata, West Bengal, India

• In this short research internship, I worked on building a ML library from scratch in Python. I primarily worked on regression, neural networks, decision trees and SVMs.