Looking for a job

Shivram Nekkanti

@shivramnekkanti

Message

I build production LLM systems, agentic workflows, and low-latency AI integrations.

United States

Message

What I'm looking for

I'm looking to build reliable, low-latency AI systems using LLMs, RAG, and agentic workflows, especially in regulated environments where privacy, evaluation rigor, and production reliability matter.

I've built production AI systems at FM Global and Northeastern University's Institute of Experiential AI, from multi-turn LLM reporting and claims-processing agents to document validation for underwriting workflows. My work has reduced retrieval latency by 300 milliseconds, cut manual review time by 75%, and supported reliable, regulated AI deployments.

I work across RAG, agentic workflows, real-time inference, evaluation frameworks, and voice systems, with hands-on experience in Temporal, vLLM, Kubernetes, LiveKit, SIP, LangChain, Redis, Celery, Docker, and Hugging Face Transformers. I also bring insurance-focused ML experience in claims, underwriting, privacy, and production reliability.

Experience

Work history, roles, and key accomplishments

Data Scientist Intern

FM Global

May 2026 - Aug 2026 (3 months)

Developed a multi-turn LLM reporting system and engineered durable agentic workflows with Temporal. Optimized vLLM inference on Kubernetes and built eval loops using LLM-as-judge methods.

LLM Application Development Prompt Architecture RAG Agentic Workflows Temporal VLLM Kubernetes LLM As Judge Statistical Measurement Telephony Systems LiveKit SIP

Data Scientist Co-op

Institute of Experiential AI, Northeastern University

Aug 2025 - Feb 2026 (6 months)

Built an LLM-assisted document-validation system and deployed deep learning models in Docker containers with CI/CD pipelines. Engineered distributed data stores and tested ETL pipelines managing 23 million records.

LLM Application Development API Integration Docker CI CD Deep Learning ETL Database Design PII Handling Security

Machine Learning Engineer

Indian Servers

Sep 2023 - Aug 2024 (11 months)

Fine-tuned intent-classification coding agents and implemented LangChain-based asynchronous inference and caching workflows. Optimized Redis and Celery task queues and enhanced DistilBERT fine-tuning.