Keep the tool behind the scenes.
I designed the assistant to evaluate an agent’s draft and produce coaching signals. It does not send a message to a homeowner, approve a refund, or make legal decisions.
An internal-facing AI QA concept for reviewing difficult homeowner-response drafts, surfacing risk, and making coaching more specific.
Paste a draft and its case context, inspect coaching notes and risk signals, and keep the final decision with the team lead. The prototype uses sample escalation scenarios.
Prototype demo · no Angi production rollout or performance result is claimed.
Explore the Angi concept
A team lead reviewing escalated homeowner responses has to spot tone issues, incomplete resolutions, and policy or legal risk without letting a generic AI voice replace human judgment. The weekly QA workflow also has a template library that can grow stale or repetitive.
I designed the assistant to evaluate an agent’s draft and produce coaching signals. It does not send a message to a homeowner, approve a refund, or make legal decisions.
The prototype takes complaint context, response draft, case type, and policy notes, then returns a rubric across tone, empathy, resolution clarity, policy accuracy, and risk, with phrase-level feedback.
A companion adoption guide proposes a small pilot, 30-day check, and review of repeat use, time saved, coaching quality, template cleanup, and missed or excessive risk flags.
Explanatory system view · reconstructed from the project
I built the interactive QA review prototype and wrote the adoption and measurement guide. The product framing, scenario templates, review workflow, and human-in-the-loop safeguards were part of my concept.
A working interview demonstration and a proposed pilot plan show how a team lead could use AI as a first-pass reviewer while retaining final control. No company rollout or measured time savings are claimed.
This is an independent interview prototype for Angi, not an Angi employment role or a deployed internal tool. The example preview contains sample scenarios; real customer information should be handled only in approved infrastructure. Pilot targets are plans, not results.
Out of the chat. Into the workflow.