Harshwardhan Fartale

ZS

• Built an agentic AI platform for life sciences, automating FDA dossier reviews with word-level traceability across 2,000+ source documents.
• Built and scaled a custom semantic LLM prompt router to handle 3,000+ requests daily.
• Post-trained small language models for healthcare, achieving performance comparable to frontier models.

DRDO & (IISC) Indian Institute of Science, Bangalore

• Wrote C++ inference modules for CASDIC combat-aircraft hardware, optimizing ML model execution under embedded latency constraints and reducing inference time by 35% versus baseline.
• Fine-tuned IBM Granite and Gemma models for scientific ML tasks, built a Graph-RAG pipeline that outperformed baselines by 45%, and led development of PULSE and SciREX under AiREX Lab.

Bajaj Finserv Health

• Built a fraud detection model that caught 20% more abusive transactions than the existing rule-based system, processing 3000+ transactions/day across Bajaj Health’s payments pipeline.
• Built a graph-based fraud detection pipeline in Neo4j and Memgraph with optimized SQL feature extraction, improving linked-fraud recall by 55% and model precision by 30% versus tabular baselines.

OpenMined

• Wrote beginner tutorials and technical documentation on differential privacy and federated learning for PySyft, and integrated them into OpenMined’s onboarding docs.
• Published blog posts explaining privacy-preserving ML concepts, driving monthly reads across OpenMined channels.

Alcuin — Deep AI Learning Platform

• Built a learning platform with mathematically rigorous AI courses and structured topic flows for many simultaneous learners.

IISc Bangalore

DRDO Industry Academia Centres of Excellence (DIA-CoEs)