Aug 2024 – Sep 2025 · 1 yr 1 mo · Pune, Maharashtra, India · On-site

• Techstack: Angular, SpringBoot, Python, SQL, .NET, LLMs, Docker, Kafka, Data Analysis
• 1. Developed Mellon-Bot, an internal tool that translates natural language queries into complex SQL statements to retrieve data from Mellon’s private databases, enhancing accessibility for non-technical users and promoting modular, self-service analytics across teams.
• 2. Contributed to a unified Tax Optimization platform by integrating features from Tax Managed Equity and Smart, streamlining backend architecture and reducing API latency using multiple methods like pagination, caching frequently accessed data, optimizing database queries and async processing.
• 3. Identified and resolved critical application vulnerabilities on the Mellon Compliance platform, strengthening its security posture, and collaborated with the Investment Management team to develop the Distribution Reporting Tracker, an automated solution for generating and distributing regulatory and performance reports, reducing manual effort and risk of errors.

Jun 2024 – Aug 2024 · 2 mos · Lausanne, Vaud, Switzerland · On-site

• 1. Working on Aiflows: https://github.com/epfl-dlab/aiflows. Specifically implemented code for running flows on your own local server.
• 2. Research Question: Would doing reward conditioning/critique generation before the model response improve the training results?
• 3. Will competitive programming questions benefit from reward conditioning before training? Can we use the critique as the feedback for code debug flow?
• More on CCflows here:https://github.com/epfl-dlab/cc_flows

Jan 2024 – Jun 2024 · 5 mos · Bengaluru, Karnataka, India · Remote

• Develop and refine components of job and resume parsers using LLMs and NLP techniques.
• Develop recommendation algorithms/engines to find ideal jobs for candidates & vice versa.
• Collaborate with cross-functional teams to integrate AI/ML components into the product.
• Research, experiment, and iterate to build and improve AI/ML/LLM models.
• Develop production APIs to deploy your AI/ML models in Python (Flask/Django/FastAPI).
• Participate in the full software development life cycle, from design to deployment.

Jun 2023 – Sep 2023 · 3 mos · Lausanne, Vaud, Switzerland · On-site

• I am currently working on sequential course recommendation system using knowledge graphs and reinforcement learning. The idea behind the work is to find different path recommendations to pursue a course.
• In general, Udemy consists of many courses contributed by random teachers, and the relationship matrix between courses and teachers is very sparse. Unlike other platforms, e.g., EdX and Coursera, where the main contributors are big universities, the content is more structured and much easier to navigate. The above makes providing a recommendation more challenging but more valuable.
• The students' participation data was extracted from students' post reviews that were submitted by the students voluntarily after they completed the course. So, the knowledge graph student-course relationship means a student completed a course and chose to fill out a review. This matrix is also very sparse, with 6-7 course ratings per student.
• We are using a Policy guided path reasoning(PGPR) approach for the knowledge graph representation and predict the state and actions on the basis of reward function.

May 2023 – Oct 2023 · 5 mos · Geneva, Switzerland · Remote

• Toolkit for Multivariate Analysis (TMVA) is a multi-purpose machine learning toolkit integrated into the ROOT scientific software framework, used in many particle physics data analysis and applications. Currently, we are developing a fast inference system in TMVA, called SOFIE, that takes takes ONNX model as input and produces compilation-ready standalone C++ scripts as output. These scripts will then provide users an easy way to deploy their deep learning models in their physics software and analysis frameworks. This project will focus on development of some missing deep learning operations which will allow to build more complex networks within TMVA for parsing the Transformer based models and Graph Net Models in SOFIE. Specifically, we propose to implement the inference functionality of some ONNX operators in the code generation format. The student can choose to build this based on existing implementations in TMVA or other existing machine learning software tools or build their own from scratch. The expected result is a working implementation of modular operators classes that implement the operators as defined by the ONNX standards in the code generation format. The project requires also to write the corresponding unit tests need to validate the written code.

Nov 2022 – May 2023 · 6 mos · Mumbai, Maharashtra, India · On-site

• The Project is under guidance of Dr. Pushpak Bhattacharyya.The idea of the research project is to create mappings of synsets among languages in IndoWordnet and linking it with Princeton’s Wordnet. Currently, manual adding/ mapping of synsets and concepts are done which is inefficient.
• The motive is to find the word embeddings of synsets including each of the words in gloss, example synset and synonyms then calculate the synset embeddings. Then we need to find the cosine similarity between the embeddings of synsets and provide automatic linking between the synsets of different languages

May 2022 – Mar 2023 · 10 mos · Remote

• Toolkit for Multivariate Analysis (TMVA) is a multi-purpose machine learning toolkit integrated into the ROOT scientific software framework, used in many particle physics data analysis and applications. Currently, we are developing a fast inference system in TMVA, called SOFIE, that takes takes ONNX model as input and produces compilation-ready standalone C++ scripts as output. These scripts will then provide users an easy way to deploy their deep learning models in their physics software and analysis frameworks. This project will focus on development of some missing deep learning operations which will allow to build more complex networks within TMVA. Specifically, we propose to implement the inference functionality of some ONNX operators in the code generation format. The student can choose to build this based on existing implementations in TMVA or other existing machine learning software tools or build their own from scratch. The expected result is a working implementation of modular operators classes that implement the operators as defined by the ONNX standards in the code generation format. The project requires also to write the corresponding unit tests need to validate the written code.