At Uniphore, I build enterprise LLM post-training systems, including long-context fine-tuning for Gemma 4 12B and 31B. I also developed the fine-tuning platform and RAFT data strategy, and designed evaluation methods to trace model failures to ingestion, retrieval, or generation.
At Axtria, I led a 10-member AI team delivering predictive and driver-analysis models for pharmaceutical clients. My work included early-adopter prediction that reached 85% accuracy and improved target identification by 30%, as well as marketing-mix modeling and an automated promotional budget optimizer.
