Atharva Deshmukh

About

Applied LLMs to high-stakes domains (wildfire risk for emergency teams and loan approval via a fine-tuned IBM Granite model), with a strong focus on reliability—using RAG-based cross-validation to reduce hallucinations and continuous ingestion pipelines (MODIS satellite imagery via AWS Lambda) to keep data current. Experienced in production orchestration and MLOps-style workflows using Airflow, AWS Step Functions, and SageMaker Pipelines, and collaborates closely with analysts on KPI-driven evaluation.

Experience

Education

Key Strengths

• Built and deployed LLM-powered conversational search for wildfire risk Q&A
• Mitigated LLM hallucinations via RAG-based cross-validation against source data
• Designed continuous ingestion pipelines (satellite imagery) with frequent vector store refreshes
• Hands-on orchestration experience across Airflow, AWS Step Functions, and SageMaker Pipelines
• Built end-to-end ETL pipeline for churn prediction with scheduled daily extraction
• Fine-tuned an LLM (IBM Granite) for loan approval prediction/generation with validation layer
• Communicates ML concepts to non-technical stakeholders using documentation and visual explanations (loss, data drift)