AI Agents
I build AI agents that do useful work in production: research, document operations, support workflows, internal tooling, and multi-step actions with clear failure modes. The focus is not on demos. The focus is reliability, observability, and operating cost.
Codex/Claude help accelerate repo understanding, workflow mapping, tool contract design, implementation slices, evaluation harnesses, and docs. I still own the actual product behavior, safety boundaries, and production acceptance criteria.
Premium, constrained ad-generation system for product marketing teams, ecommerce brands, app founders, and agencies. Focused on a single high-quality workflow from brief to canonical export and delivery.
Production-grade deep research agent that plans, fetches, and writes structured reports from URLs and documents with strict guardrails.
DocOps automation agent for production-grade document ingestion, grounded Q&A with citations/refusals, and an audit harness for evaluation.
Yes. That is the normal use case. I design the agent around explicit tool contracts, auth boundaries, retries, and failure handling so it can operate safely against your internal APIs and workflows.
Production systems. If a prototype is needed first, it is designed to reduce risk on the production path, not to create throwaway code that has to be rewritten immediately after validation.