Integration Design
Connecting separate source records into one reliable view without hiding where each answer came from.
- API integrations and fixture-backed adapters
- SQLite-backed sample data models
- Source citations and traceable records
BACKGROUND
Systems that connect records, explain outputs, and leave room for judgment.
I like software where the useful work is in the boundaries: pulling together records from different sources, validating what an agent proposes, and designing review surfaces that make decisions easier to trust.
Working Range
Production-Shaped Builds
Building demos with the boundaries I would want in a real service: typed contracts, tests, local fixtures, and deployment paths.
- Serverless functions and queue-shaped flows
- Webhook-style integrations and mock channels
- Readable setup docs and design notes
AI System Design
Using LLMs for retrieval, drafting, and routing while keeping source data, code validation, and human approval visible.
- Tool use over explicit data sources
- Schema validation outside the model
- Grounded answers with inspectable evidence
Review Interfaces
Designing screens that show what changed, why it matters, and what still needs a person to approve.
- Dashboards and analytics summaries
- Human-in-the-loop approval states
- Per-entity history and audit trails
How I Think About Tools
01
Start from records. Good demos still need real-shaped inputs, constraints, and failure modes.
02
Make outputs inspectable. Every generated answer should point back to data the user can check.
03
Keep approval explicit. AI can propose and summarize; code validates; people decide when the stakes call for it.