Experience
Research Scientist
NielsenIQ · India
Jul 2026 – Present
Core Models
- Part of the Core Models team, under David Jimenez.
Applied Research Scientist
Thomson Reuters Lab · India
Aug 2024 – Jun 2026
Legal AI Reasoning and Model Enhancement
- Pioneered a legal synthetic data generation pipeline for Process Reward Models (PRMs), creating domain-specific training datasets that improved legal reasoning capabilities on LegalBench benchmark tasks.
- Joint first author of Test-Time Scaling in the Wild (NeurIPS 2026), the first compute-normalised comparison of test-time scaling on open-ended tasks. We compared five TTS families across five benchmarks (medicine, law, finance, general chat, creative writing). We showed that exploitation, not exploration, is the bottleneck. Reward models correlate at only ρ_v ≈ 0.12 with true quality, so selecting the best candidate is close to random at any budget.
- Architected an IRAC (Issue-Rule-Application-Conclusion) Knowledge Graph framework using Thomson Reuters’ Westlaw corpus and court case data, generating high-quality preference datasets that improved legal reasoning alignment in fine-tuned LLMs.
AI-Powered Legal Document Update System
- Built a comprehensive end-to-end LLM workflow pipeline to update Word documents with XML parsing and intelligent contextual mapping, achieving 70%+ success rate while preserving complete document formatting integrity.
- Engineered a human-in-the-loop validation interface with reasoning chains and alert-point mapping, resulting in a 60% reduction in manual processing time while maintaining legal accuracy.
DocEvolver
- Created an MVP for a “Cursor for Word” style extension for updating and understanding MS Word files for lawyer-editors.
Search-and-Replace Agentic System
- Architected a multi-agent AI system for automated Word editing with a comprehensive validation pipeline (schema enforcement, content integrity, audit logging), achieving 98% accuracy and 65% reduction in manual content revision.
- Constructed error-resolving agents with function calling and multi-turn reasoning to fix XML issues, implementing few-shot learning and self-healing mechanisms — adopted across teams, processing hundreds of documents monthly with sub-8 second processing time per section.
Additional Tools
- Truth Social Monitor: Created a monitoring system for Trump’s Truth Social posts with sub-3 second latency, generating automated alerts for Reuters journalists.
- Page Flipper: Revived the Page Monitor extension for website tracking, eliminating Visual Ping subscriptions for the team.
Research Intern
Microsoft Research India
Jan 2024 – Jul 2024
Programming with Representations (PwR)
- Led backend development for PwR Studio, focusing on Natural Language to Domain Specific Language (NL2DSL) translation using GPT-3.5 and GPT-4.
- Developed a symbolic translation pipeline that generates finite state machines structured as custom DSL, achieving an 85% reduction in hallucinations.
- Formulated rubrics and evaluation loops with error correction over DSL, increasing valid DSL generation from 65% to 95%.
Jugalbandi Studio Engine
- Architected a Python-based platform that converts DSL into scalable finite-state-machine-based chatbot applications, reducing development time by 80%.
- Platform was featured in Satya Nadella’s keynote talks; enabled 15+ non-technical organizations to build AI-powered conversational bots.
Jugalbandi (JB) Manager
- Established a chatbot management platform supporting WhatsApp, Telegram, and Web channels with multilingual text and voice capabilities.
- Integrated Bhashini Speech models with Azure service failover mechanisms, enabling 70% faster deployment of new chatbots.
Research Intern
AI Institute, University of South Carolina
Dec 2022 – Apr 2024
Knowledge Enabled Multimodal Ingredient Substitution
- Master’s thesis on multimodal ingredient substitution. Built a knowledge graph incorporating 27K ingredients and 80K substitution pairs, enabling precise ingredient recommendations using multimodal and constraint-based search.
- Developed an LLM-based query module for the ingredient substitution knowledge graph.
- Formulated cross-modal recipe retrieval and cooking action recognition for recipe analysis, achieving 95% recall and leading to Cook-Gen at IEEE SMC 2023.
Visiting Researcher
Societal Computing, Saarland University (SIC)
May 2023 – Aug 2023
Satellite Image Super-Resolution
- Worked on satellite image super-resolution using temporal and multispectral information.
- Leveraged high temporal frequency low-resolution data for wildlife tracking and improved disaster analysis through GAN and diffusion approaches.
Research Intern
University of Maryland, Baltimore
Oct 2022 – Apr 2023
Personalized AI Assistant (HealthCareNLP Grant)
- Developed personalized response generation models using reward scaling over BART and T5.
- Work accepted as K-PERM at AAAI Symposium 2024; improved NUBIA score by 10%.
- Focused on knowledge and persona-aware loss scaling for better response generation.
AI Intern
EdgeNeural.ai · Pune, India
Jun 2022 – Aug 2022
Accelerated Inference & Model Optimization
- Developed training and optimization pipelines for automatic model training and hosting.
- Accelerated inference through quantization and CPU/GPU customization using TensorRT and OpenVINO.
Research Intern
Video Analytics Lab, IISc Bangalore
May 2022 – Aug 2022
StyleGAN-based Video Analysis
- Implemented StyleGAN-based architectures for disentangled video interpretation across domains.
- Improved image/video generation and analysis workflows using GAN-based modeling.
Research Intern
Visual Learning and Intelligence Lab, IIT Hyderabad
Nov 2021 – Apr 2022
Medical Image Processing
- Researched medical image processing with Prof. Dr. C. Krishna Mohan.
- Developed a novel architecture for improved classification of low-quality images and unbalanced datasets.
- Published SFFNet for panoramic dental X-ray segmentation at IEEE APSCON 2023.
Computer Vision Engineer
AI Mage (WETHEKOO)
Mar 2021 – Apr 2021
Fashion Tagging Engine
- Developed a fashion tagging engine using deep learning for product categorization.
- Optimized and deployed computer vision models on edge devices.
Software Engineer
Rhizicube Technologies
Jun 2021 – Sep 2021
Consumer Data Platform
- Oversaw server and REST API development and database design using Golang (Gin).
- Built a real-time streaming data pipeline using Apache Kafka.
- Built a LinkedIn scraper and generalized organization website crawler using Selenium and Beautiful Soup.