Jun 2025 – Aug 2025 · 2 mos · New Jersey, United States · On-site

• Document AI

Jan 2024 – Aug 2024 · 7 mos · Bengaluru, Karnataka, India

• ● Built a Conversational AI bot for health insurance domain by leveraging RAG, multi agentic workflows and LLMs - gpt-4o.
• ● Designed scalable RESTful APIs integrating tool calling, intent classifier, and multi-turn query management with Redis caching
• ● Deployed production-grade LLM workflows using Jenkins, ECR, Lambda, Bedrock and DynamoDB to serve high scalability.
• ● Utilised PEFT finetuning, reranking & prompt engineering techniques like CoT, CoVe, React to mitigate hallucinations by 25%.
• ● Modulated model evaluation cron jobs using LLM as a judge with metrics like faithfulness, context relevance etc.
• ● Worked on API observability using AWS CloudWatch to monitor latency, throughput, and error rates of inference endpoints.

May 2023 – Jan 2024 · 8 mos · Bengaluru, Karnataka, India · On-site

• ● Led a team of five data scientists and contributed individually to the development of clinical healthcare products.
• ● Architected modular Airflow DAGs for clinical temporal summarization using Patient’s EHRs by finetuning BERT for NER,
• sentiment detection via MLM & association extraction (0.92 F1 Score) on Pytorch.
• ● Configured SLA monitors & failure alerts for DAG tasks to track EHR delays, model output integrity, and retrigger failed steps.
• ● Engineered critical components including OCR (pytesseract), document classification models & section header detection.
• ● Built a Graph RAG based chatbot for Q&A on medical contracts utilising Neo4j, Langchain & code generative LLMs

Apr 2022 – May 2023 · 1 yr 1 mo · Bengaluru, Karnataka, India · On-site

• ● Worked on training and deploying BERT-based sentiment, association, and NER models on patient EHRs stored in MongoDB.
• ● Optimized BERT inference pipeline using ONNX Runtime; achieved real-time performance (0.001 sec/sentence) on GPU.
• ● Leveraged NLP augmentation libraries to enrich training data, achieving a 3% F1 score improvement (89% to 92%).
• ● Prototyped cardiovascular disease detection via multimodal infusion of 1D ECG and 2D spectrogram using ResNet/AlexNet.
• ● Built transformation of medical codes like ICD10, Snowmed, Lionc by fine tuning SentenceBERT on contrastive loss

Apr 2019 – Apr 2022 · 3 yrs · Bangalore Urban, Karnataka, India

• ● Led AML/anti-fraud typology repository management initiatives for financial institutions .
• ● Developed Pyspark based CI/CD pipelines for ML lifecycle (training - deployment - monitoring) using Airflow, Docker.
• ● Developed and deployed anomaly scoring models (IF trees, Z-score), XGBoost, and clustering models (KMeans++, DBSCAN) in
• both unsupervised and supervised AML pipelines.
• ● Worked on run rate and affordability based collection algorithms for their credit card payment platform.
• ● Executed credit risk modeling and feature engineering using CatBoost and derived features from credit report data.

May 2017 – Apr 2019 · 1 yr 11 mos · Bengaluru Area, India

• Led a team of 5 OTA sales specialists to help them take data driven decisions for increasing hotel bookings.
• Developed a Ranking Recommendation System for Hotels listed at In house platform.
• Predicting the cancelation probability of customer Bookings using RF Trees(realization improvement by 10%).
• Worked on SQL metabase dashboards to track various northstar metrics.