Niranjani Pandey

Directed design of a two-stage safety architecture for harm detection in youth mental health dialogue, achieving F1 = 0.931 with 93.8% recall. Led evaluation strategy and built a synthetic data engine generating 10,471 quality-scored conversations.

Built a multi-agent text-to-SQL assistant using CrewAI and Gemini, lifting answer accuracy from ~13% to ~67%.

Built an end-to-end pipeline turning satellite imagery into transparent, auditable scores for regenerative farming practices, with 90+ features per parcel and 221 automated tests.

Leads analytics and AI for Avion Rewards, building KPI frameworks, multi-agent AI reporting, and dashboards that drive executive decisions. Delivered analytics for partnerships bringing in 30,000+ new members and managed a team of analysts.

Architected digital analytics data foundation for 12 e-commerce properties, reducing compute costs by $350K+ and improving decision-making efficiency by 40%. Drove customer engagement analytics and feature development that increased average order value by 35% and conversion by 15%.

Developed and benchmarked unsupervised clustering and transformer-based models for human trafficking detection, improving evaluation signal quality and model accuracy by 14% through data augmentation.

Applied program covering mechanistic interpretability, LLM safety, and agentic system design.

AI project lifecycle management, governance frameworks, and stakeholder alignment for AI initiatives.

Graduate-level program covering probability, machine learning, and data analysis.

Quantitative foundation in statistical inference, mathematical modeling, and economic analysis.