Ragamalika Karumuri

About

AI Engineer at Humanitarian AI who has built and productionized both a LangGraph-based multi-agent workflow system and a RAG pipeline (OpenAI embeddings + vector DB) with rigorous evaluation/guardrails. Reports strong measurable impact (60% faster workflow delivery, 40% fewer incidents, 70% reduced research time) and has prior enterprise modernization experience at Infosys migrating ETL to microservices with zero production incidents.

Experience

Education

Key Strengths

• Built and shipped an end-to-end LangGraph multi-agent workflow system (frontend, backend, data, infra) to production in <6 weeks
• Delivered measurable ops impact: ~60% faster workflow delivery and ~40% fewer production incidents
• Productionized RAG with OpenAI embeddings + vector DB; improved research time ~70% with source-accurate answers
• Strong evaluation/guardrails mindset (accuracy/relevance/task completion scoring, faithfulness checks, latency testing, alerting) enabling shipping without human-in-the-loop review
• Legacy modernization leadership: refactored ETL into microservices via parallel runs, feature flags, incremental cutover with zero production incidents
• Effective stakeholder translation (vague CPO goal to prioritized roadmap; shipped smaller MVP first, iterated from user feedback)
• High-autonomy execution across parallel projects with minimal oversight; timeboxed work and shipped working code quickly
• Productionizes LLM prototypes with modular stages, standardized schemas, and monitoring
• Improves reliability via soft validation, retries, and fallback prompts to handle real-world inputs
• Uses measurable SLO-style thresholds (latency, accuracy, failure rates) to ensure quality
• Systematic real-time debugging using logs/traces/metrics and reproduction with real inputs
• Collaborates transparently with customers to validate fixes quickly and build trust
• Delivers effective developer workshops (hands-on RAG agent to production workflow progression)
• Supports deal closure and adoption by building tailored PoCs and live demos addressing reliability/scalability concerns