Lead Data Scientist
Data Scientist
Available for hire
Years of
experience
6+ years
Experience
level
Senior
Available
for
Full-time, Part-time, Contract, Freelance
Available from
07 Oct 2026
Data Scientist with 6+ years of experience building and deploying production AI systems across Computer Vision, NLP , and Generative AI. Specialized in multimodal AI, document intelligence, RAG, and end-to-end ML systems, with experience delivering measurable improvements in automation, accuracy, and inference performance.
Languages
No languages.
Employment History
Lead Data Scientist at
Roadzen Inc
Current
2025
-
Now
o Built Autospect, a Claude-powered motor insurance claims agent that automated 80% of the pre-inspection workflow, reducing claim intake turnaround time from hours to minutes.
o Engineered a from-scratch instance segmentation pipeline using a frozen DINOv2 backbone and Mask2Former head for car-part segmentation, achieving ~93% AP50 on the test dataset.
o Fine-tuned a 4-bit quantised Qwen-vision 2.5vl with LoRA, improving extraction accuracy by 37%, with key business fields achieving up to 90% accuracy.
o Created a Gemini-based vehicle identity extraction pipeline with 97.8% accuracy.
Research Assistant at
University of Edinburgh
2024
-
2024
o Scraped and processed 70GB of historical FTSE All-Share annual reports for financial QA research.
o Benchmarked GPT against 4-bit quantized Llama and Mistral models on QA tasks, with GPT outperforming both by ~12% in answer accuracy.
o Improved QA performance by 16% over baseline RAG pipelines by implementing a MapReduce-based ranking architecture with comparable inference latency.
Senior Data Scientist at
Roadzen Inc
2020
-
2023
o Built end-to-end ML pipelines for data ingestion, training, and model serving using FastAPI, Airflow, and Jenkins, enabling automated retraining and deployment of in-house ML models.
o Enhanced an insurance policy QA system using GPT-based retrieval, achieving 95% answer accuracy.
o Built a Mask R-CNN model for instance segmentation of vehicle damage, parts, and profiles using PyTorch, reducing claims processing time from 40 minutes to under 2 minutes.
o Redesigned the mAP evaluation metric to better account for annotation subjectivity, improving model performance achieving 74 mAP.
o Reduced Damage Recognition API inference latency by 30% using TorchServe and FastAPI.
Data Scientist at
Spoonshot
2019
-
2020
o Designed a weighted DeepWalk-based graph model using ingredient co-occurrence and flavour pairing theory to generate novel flavour combinations.
o Scaled graph edge weight generation to over 100 million ingredient combinations using PySpark.
o Implemented a Fast R-CNN pipeline for nutrition panel extraction from product images, achieving 85 mAP.
Education
Msc in Data Science at
University of Edingburgh
2023
-
2024
Btech in IT and Mathematics at
University of Delhi
2015
-
2019