
Applied Science Intern @ JP Morgan ML Center of Excellence
Intern
- Developing a robust evaluation system for News RAG Pipelines.
- Improved robustness and human alignment over open-source baselines and LLM-as-a-judge frameworks.

Data & Applied Scientist @ Microsoft
Full-Time
- Built LLM-based pipelines in Bing to improve user query understanding and deliver more relevant answers.
- Trained and optimized XLMR-based automated entity extractor on 20B Bing webpages (1.25ms latency).
- Bootstrapped data refresh and triggering pipeline: yielded 1.5M DAU gain (2x YoY).
- Published SegRank, a novel LLM grounding method reducing hallucinations by 15.7% in BingChat.
- Honorable Mention at the Executive Challenge in the Microsoft Global Hackathon 2024
- Won the Award of Excellence for Innovation in Microsoft, IDC for my work on the events datapipeline.
- Fast-tracked to L60 at Microsoft, earning promotion in 9 months due to exceptional ownership and a proven track record of delivering impactful project outcomes.

Student Researcher @ ETH Zurich & MPI Tübingen
Part-Time
- Formulated CausalCite, a novel causal metric for paper citations.
- Ran large-scale experiments on 2B+ edge and 200M+ node citation graphs.
- Proposed evaluation metric showing 30.14% better correlation with test-of-time than citation count.
- Accepted at ACL 2024 (Main Conference), Bangkok; presented as first author.

ML Engineering Intern @ Deloitte AI Center of Excellence
Intern
- Modelled probability of success for in-bound opportunities, improving prioritization by 76%.
- Built an internal MLOps platform for converting notebooks to Kubeflow Pipelines.

ML Engineering Intern @ Zomato
Intern
- Detected fraud in new user accounts (35% of fraud cases) using ML models.
- Trained seq2seq models on clickstream data to learn user behavior patterns indicative of fraud.

Research Intern @ Video Analytics Lab, IISc Bangalore
Intern
- Built a self-supervised pipeline for generating pose and 3D colored mesh from 2D RGB images.
- Explored VAEs, AAEs, and Hierarchical AAEs as priors for pose estimation.
- Used PyTorch3D for differentiable rendering; processed Freihand, HUMBI, and HO3D datasets.

Research Intern @ Complex Networks Research Group, IIT Kharagpur
Intern
- Developed a QA model on the AmazonQA dataset using SentBERT for review filtering.
- Benchmarked seq2seq and retrieval-based models (including RAG) in PyTorch.