Karthik Talluri

Designed and deployed production-grade LLM guardrails using NVIDIA NeMo, TruLens, and transformer-based classifiers, improving bias/toxicity detection precision by 10%. Built full-stack AI safety infrastructure with Dockerized FastAPI services and AWS Lambda optimization, improving reproducibility and inference efficiency by 15%. Led development of fairness and carbon impact metrics, contributing to stakeholder alignment and increasing technical buy-in during investor discussions. Contributed to small language model distillation, groundedness evaluation, and prompt risk assessments improving factual accuracy of outputs by 10% and enabling scalable deployment across internal tools.

Built and maintained scalable dbt pipelines to transform Gong and HubSpot data into analytics-ready models, reducing ETL latency and pipeline failures by 30%. Designed dimensional data models integrated with a vector database to power LLM-driven semantic search, improving insight generation and retrieval precision by 20%. Partnered with design partners to validate AI workflows and surface key GTM signals, enhancing cross-source alignment and reducing manual triangulation.

Built Streamlit-based (RAG) prototypes and interactive demos integrating TruLens and NeMo Guardrails, enhancing investor engagement and reducing post-demo drop-off by 20%. Optimized Dockerized FastAPI pipelines for metric-serving, reducing image size and cold-start latency for AWS Lambda deployment by 30%. Collaborated on early-stage experimentation workflows and contributed to rapid prototyping used in pitch materials.