Back to Projects

Apple · 2021 – 2026

An agentic AI application that assesses and migrates lakehouse jobs through a conversational interface

MCP servers and a chat interface so teams can assess and migrate Snowflake ingestion jobs to the lakehouse by talking to an agent, plus evals so I could trust what it said.

Senior Software Engineer, Data Platform

Assessment time

Weeks → min

Agent tooling

MCP

Reliability checks

LLM evals

Problem

To find out if an ingestion job could move from Snowflake to the lakehouse, someone had to dig through it by hand. One assessment could take weeks.

Approach

  • Built MCP servers that give the agent the lakehouse context it needs to judge whether a job can migrate.
  • Built a full-stack chat interface so teams can assess and migrate jobs just by talking to the agent.
  • Built an LLM evaluation framework to check the agent was reliable before anyone depended on it.

Results

  • Assessments went from weeks to minutes.
  • The agent’s answers are checked by evals, not taken on trust.

Stack

  • MCP servers
  • LLM agent
  • LLM eval framework
  • Python
  • Full-stack web app
  • Snowflake
  • Apache Iceberg