Rukmini Pisipati

About

Built and shipped a production LLM hallucination detection and monitoring pipeline using semantic-level entropy (embedding-clustered multi-generation variance) to flag unreliable outputs in downstream automation. Implemented a scalable async architecture (FastAPI + Docker + Redis/Celery) with strong observability (structured logs + PostgreSQL) and developed evaluation loops combining controlled prompts and human review; also partnered with non-technical stakeholders on AI-driven form validation/document processing.

Experience

Education

Key Strengths

• Built and deployed production LLM hallucination detection/monitoring using semantic entropy
• Designed scalable async processing architecture with Redis queues and Celery workers
• Strong observability practices (structured logging + PostgreSQL analytics/monitoring)
• Pragmatic evaluation strategy for ambiguous labels (controlled prompts + human review)
• Reliability-focused AI workflow design balancing correctness, latency, and cost
• Effective collaboration with non-technical stakeholders using examples, dashboards, and threshold alignment

Reference Highlights

• Wrote strong foundational code
• Foundational contributions to algorithm components
• Valuable contributor on a small team
• Collaborative in design discussions
• Adaptable across multiple parts of the stack
• Bridges technical and non-technical communication effectively
• Improved communication over time
• Able to focus messaging on high-impact details for stakeholders