I’m currently working on
Operationalizing a self served ML platform that spans the full model lifecycle from experimentation to production. This includes Valohai and Anyscale for scalable training and distributed compute, MLflow for experiment tracking and governance, Seldon Core v2 for model serving, Qdrant for vector search, and Airflow for orchestration. In parallel, I am architecting GenAI infrastructure with a strong emphasis on agentic workflows, LLM gateways, and a unified synthetic data generation framework built around the Model Context Protocol (MCP).

I’m looking to collaborate on
Designing and hardening production grade GenAI platforms, particularly around agent orchestration, tool calling, retrieval augmented generation, evaluation frameworks, and scalable inference patterns. I am especially interested in collaboration at the intersection of data platforms, ML platforms, and GenAI system design.

I’m looking for help with
Real world benchmarks, failure modes, and operational lessons from large scale agentic systems, LLM gateways, and vector databases in production. I am also keen to exchange ideas on standardizing MCP based integrations across heterogeneous model and tool ecosystems.

I’m currently learning
Advanced agentic architectures, LLM routing and governance patterns, synthetic data generation for training and evaluation, and the emerging standards shaping interoperable AI systems, including MCP and open inference protocols.

Ask me about
Building self served ML platforms, MLOps at scale, GenAI infrastructure, agentic workflows, LLM gateways, vector search, data platform architecture, and the practical tradeoffs of running AI systems in production.

Fun fact
I have watched data engineering evolve from hand rolled ETL and on prem warehouses to streaming platforms, ML platforms, and now agent driven AI systems, and I still enjoy debugging pipelines more than writing slide decks.