- Design, build, and operate production AI agents, including an autonomous coding agent that works from Jira and Slack, opens pull requests, fixes CI failures, and applies review feedback.
- Build the retrieval and context layer using embeddings, semantic search, knowledge-graph modeling, Snowflake semantic-layer integration, and internal documentation.
- Extend the internal MCP server with reusable skills, tools, and processes, and help product teams integrate agents into their workflows.
- Build offline test sets and online metrics to evaluate prompts, models, and retrieval changes, and gate releases on evaluation results.
- Reduce cost and latency by routing work to suitable models, using caching and structured outputs, and tracking spend per agent and task on AWS Bedrock.
- Harden agents against prompt injection, memory poisoning, and over-broad permissions using least-privilege tool access, auditable actions, and human approval where warranted.
- Instrument agents with tracing and logging, diagnose failures, and participate in on-call for owned systems.
- Evaluate new models and tools and adopt those that prove effective.
- Write design documents, review code, and mentor engineers on building with LLMs.
PythonTypeScript