Kiran Kumar

About

Software engineer (4+ years) with hands-on production GenAI experience: built an AI incident triage assistant that summarizes production logs for on-call engineers and iterated it using real incident metrics (time-to-signal, triage duration). Also shipped a RAG-based customer support knowledge assistant using embeddings + vector retrieval with strong guardrails (relevance thresholds/abstain, sanitization, auditing) and a formal eval loop (500-query gold set) that drove measurable retrieval improvements.

Experience

Education

Key Strengths

• Shipped production LLM tools for incident triage with clear safety boundaries (decision support vs remediation)
• Strong tradeoff management across latency, cost, and quality (token limits, batching, filtering, prioritization)
• Data-driven iteration post-launch using operational metrics (time-to-signal, triage duration, adoption/usage)
• Built RAG-based support knowledge assistant with grounded answers and source citations
• Implemented robust LLM guardrails (relevance thresholds, abstain behavior, content filters, query sanitization, auditing)
• Designed and ran LLM evaluation loops (500-query gold set; measured hallucination/citation/relevance/latency; regression prevention)
• Improved retrieval quality via metadata filtering and query expansion (+15% relevance)
• Built scalable log ingestion pipelines with reliability and observability (retries, schema validation, DLQ, dashboards)