AI assistants / workflow products
Turn an AI workflow into a product people can control
I design the source, action, human-review, permission, and recovery states around model output—then prototype the interaction and stay close to implementation.
Primary product proof
An assistant built around real work
A private 2026 product connects Slack, Google, and YouTrack through an n8n and Node.js workflow. The available evidence verifies the delivered states and system structure; usage impact was not measured.
Workflow check
Four things the interface must make clear
The product has to remain understandable after the first successful prompt.
- 01 / Context
Sources and permissions
Show what the system knows, where the information came from, and what each user is allowed to access.
- 02 / Control
Actions and human review
Separate suggestions from executable actions and place confirmation or review where the consequence requires it.
- 03 / Trust
Failure and recovery
Design partial results, missing context, blocked actions, retries, and handoff instead of treating failure as one error message.
- 04 / Delivery
Prototype and release
Validate the interaction in a working prototype, then keep the design decisions intact through implementation and QA.
Supporting evidence
More shipped work
Media Product / Match Data + Admin
NextSport.io
Delivered from 28 January to 3 June 2026 across 139 live route nodes. Early 2026 product tests ran for at least one month; the strongest selected segment reached 15% peak click-through into match pages and deeper team, player, or tournament context. Registration, revenue, and long-term retention were not measured.