I build
AI Engineer with 3+ years of experience designing and shipping production LLM systems, RAG pipelines, and agentic AI workflows. From data ingestion and feature engineering to LLM evaluation and cloud-native MLOps — I turn complex AI research into reliable, scalable systems.
I'm Yajurved Jayavarapu, an AI Engineer based in San Francisco, CA with a Master of Science in Data Science from the University of Alabama at Birmingham. I specialise in LLM systems, RAG architectures, and agentic AI — building production-grade pipelines that go far beyond notebooks.
My work spans Syneos Health, UAB, and Thomson Reuters — where I architected Spark-based ML pipelines, reduced data latency by 35%, cut manual reporting time by 60%, and deployed NLP systems at scale. I hold deep expertise in LangChain, LlamaIndex, LangGraph, vector databases, and AWS cloud-native ML platforms.
From LLM orchestration frameworks to distributed data engineering — the full stack of modern AI.
Production-grade AI systems — from grounded RAG to multi-agent verification.
Built a grounded RAG system that enforces document-only answers and safe refusals to eliminate hallucinations. Features agent-based query routing, hybrid retrieval (BM25 + dense vectors), metadata-aware filtering, chunk-level traceability, and integration tests for end-to-end LLM validation.
End-to-end health data analytics pipeline ingesting Apple Watch metrics, applying time-series analysis and anomaly detection, and surfacing insights through an interactive analytics dashboard. Demonstrated 40% improvement in data reliability metrics.
Multi-agent RAG system with dedicated Planner, Retriever, Answer, and Verifier agents. Implements query decomposition, multi-hop retrieval, citation enforcement, and retry logic using modular state machines and structured JSON outputs.
Built Spark-based ML pipelines on Hadoop clusters for large-scale text classification and sentiment analysis at Thomson Reuters. Applied cross-validation and robust metrics (AUC, F1, Precision, Recall) with reproducible training and evaluation workflows.
Orchestrated automated ETL and ML workflows using Airflow DAGs on AWS at Syneos Health. Built reusable ML components consumed by multiple downstream analytics and AI workflows, reducing data processing latency by 35%.
Engineered real-time data ingestion pipelines from multiple sensor and API sources at UAB, improving data accuracy by 25%. Built interactive dashboards backed by standardized datasets, with anomaly detection and unified analytics layer for downstream modeling.
Whether you have a full-time AI/ML opportunity, a research collaboration, or want to discuss LLMs and RAG systems — my inbox is always open. Based in San Francisco, open to remote.