CV

M.S. in Data Science — University of Rochester  ·  Rochester, NY

Bachelor of Technology — Indian Institute of Technology, Roorkee  ·  Roorkee, IN

Head of AI — DreamStreet  ·  Mumbai

• Designed a compliance-aware orchestration framework for multi-step LLM agents in a SEBI-regulated financial setting — constrained tool-use, guardrail/policy enforcement, and end-to-end audit logging for controllability and traceability of agent decisions.
• Deployed self-hosted open-source language models (Qwen 3.5 / 3.6 27B) with vLLM and an agent-tooling stack across on-prem, GCP, and AWS.
• Built Hermes-based self-serve agents for in-house productivity — domain research reviews, alerts, and similar tasks — across Marketing, Finance, IT, Tech, and HR.
• Built full-stack AI for investor and trader workflows across research, brokerage, and advisory, and drove organization-wide AI adoption through training, rapid prototypes, and agent-centric workflow redesign.

Senior Principal Research Scientist / Head of Applied Research — Dream11 (Dream Sports)  ·  Mumbai & New York

• Headed applied AI research for Dream Sports and built a high-performing cross-continent team of research scientists, applied scientists, and ML engineers across India and New York.
• Established and led an industry–academia research partnership and research center with Columbia University, NY — co-defining research agendas at the intersection of ML and sports analytics and supporting joint work with faculty, post-docs, and PhD students.
• Defined a multi-year applied-research portfolio — ~23 candidate research problems, 10 funded over two years — spanning sports robotics, LLM-based persona/behavior simulators, and agentic evaluators for personalization.
• Developed deep-learning models for early churn prediction from large-scale user–product interaction time series (ICMLA 2023) and LLM-based behavioral simulation of user lifetime trajectories at population scale (250M+ users).
• Built distributed recommendation, semantic similarity search over ~100M entities, and feature infrastructure for 250M+ users — seeding publications on personalization (UMAP ‘25, ECML-PKDD ‘25) and entity resolution (ACL 2026).
• Designed a real-time probabilistic forecasting system producing 50k+ concurrent forecasts under strict latency constraints for production ML applications.
• Delivered regular data, ML, and AI training across Dream Sports (audiences of ~10 to 200) and co-led Sports × AI sessions at Columbia for students, post-docs, and faculty.

Staff Data Scientist — David H. Smith Center for Vaccine Biology and Immunology, University of Rochester  ·  Rochester, NY

• Built automated, self-serve ML systems for bio-imaging research, including 3D reconstruction from hyperspectral microscopy — contributing computational methods to immunology research published in Cell Reports and the Journal of Immunology.
• Automated ultrasound Doppler-angle estimation using deep neural networks (EMBC 2019).

Data Scientist — AXA Insurance  ·  Pune

Built survival / mortality-forecasting models and scaled statistical pipelines (Spark, Python) for the US population (~300M+ entities).

Data Analyst — AbsolutData Research & Analytics  ·  Gurgaon

Developed a multi-stage predictive-maintenance model fusing sensor telemetry, oil-analysis, and human-labeled alerts.

Structure-Guided Entity Resolution: Fine-Tuning LLMs for Robust Name Matching in Complex Linguistic Contexts. Chourasia, A., Kapoor, S., & Patil, N. — Association for Computational Linguistics (ACL), Industry Track. openreview.net/forum?id=rLisRb1T1Y

Optimizing Fantasy Sports Team Selection with Deep Reinforcement Learning. Proceedings of CODS-COMAD ‘24, ACM, 284–291. doi.org/10.1145/3703323.3703743

Early Churn Prediction from Large Scale User-Product Interaction Time Series. 2023 International Conference on Machine Learning and Applications (ICMLA), IEEE. doi.org/10.1109/ICMLA58977.2023.00314

CXCL10+ Perivascular Clusters Nucleate Th1 Cell Tissue Entry and Activation in the Inflamed Skin. Journal of Immunology. jimmunol.org/content/204/1_Supplement/220.9

CXCL10+ Peripheral Activation Niches Couple Preferred Sites of Th1 Entry with Optimal APC Encounter. Cell Reports / preprint at biorxiv.org/content/10.1101/2020.10.04.324525v1

Automated Ultrasound Doppler Angle Estimation Using Deep Learning. 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), pp. 28–31, IEEE. pubmed.ncbi.nlm.nih.gov/31945837

Driving Engagement in Daily Fantasy Sports with a Scalable and Urgency-Aware Ranking Engine. Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 40, No. 47. doi.org/10.1609/aaai.v40i47.41479

Enhancing Personalisation in Fantasy Sports with Graph-Based Representations. UMAP ‘25, ACM, 361–363. doi.org/10.1145/3699682.3730974

Personalized Contest Recommendation in Fantasy Sports. ECML PKDD 2025, Applied Data Science Track, Part IX, Springer, 22–35. doi.org/10.1007/978-3-032-06118-8_2

ForeCal: Random Forest-based Calibration for DNNs. CODS-COMAD ‘24, ACM, 44–51. doi.org/10.1145/3703323.3703330

FENCE: Fairplay Ensuring Network Chain Entity for Real-Time Multiple ID Detection at Scale in Fantasy Sports. AIMLSystems ‘23, ACM, Article 26. doi.org/10.1145/3639856.3639882

Maximizing Success Rate of Payment Routing using Non-stationary Bandits. AIMLSystems ‘23, ACM, Article 27. doi.org/10.1145/3639856.3639883

Accelerating Causal Algorithms for Industrial-scale Data: A Distributed Computing Approach with Ray. AIMLSystems ‘23, ACM, Article 24. doi.org/10.1145/3639856.3639880