Zero-Touch Coding for Conversational MLOPs

There are several main challenges to achieve Zero-Touch coding of MLOPs with the help of LLM as the reasoning engine. Central to achieve this objective is the use of LLM to translate the natural language prompts into domain specific instructions and execute them through MCP. To develop intelligence of the facility, we use the paradigm of treating everything as data including micro-services, hardware resources, observability metrics, databases, configuration files, workflows, scripts, codes, and all other things that are part of the experimental setup. Using the talk to data paradigm, there are three challenges that we are trying to address by developing conversational MLOPs with MCP tooling support for each of these resources: (1) talk to data for insights, reports, aggregation, summaries; (2) talk to data for MLOPs workflow; and (3) talk to data for workload management, model performance tracking, reporting incidents and anomalies. In this presentation, we will showcase our initial efforts and future activities.