Miko Planas

I build production-grade machine learning and LLM systems with a focus on multi-agent architectures, retrieval-augmented generation, and rigorous evaluation. I prioritize reliability, interpretability, and measurable business impact in AI deployments.

At Thinking Machines I led development of a multi-agent sales platform for a major banking client using PydanticAI, Databricks MosaicVectorSearch, and AWS Bedrock. I also constructed 10 RAG pipelines that achieved >90% on ROUGE-based metrics and implemented comprehensive LLM-as-judge evaluation suites with DeepEval, Pytest, and GitHub Actions.

I have provided AI ecosystem and digital strategy consulting to clients including IDEO and large real-estate and retail conglomerates, and analyzed six large-scale Snowflake datasets (10–350M rows) to surface behavioral patterns and deliver a customer-segmentation POC using HDBSCAN. I communicate technical results through executive presentations that drive revenue and product adoption.

Beyond engineering, I curate and summarize research as an associate editor, producing 200+ articles on LLMOps, MLOps, AI safety, and deep learning for 100K+ subscribers. I’m motivated by alignment, interpretability, and building responsible, production-ready AI systems in collaborative teams.