Evidence QA v1: Conversation Reviews with Checked Quotes

Preparing conversation reviews takes repeated work: search a transcript, apply the scorecard, find examples, add the points, and copy the draft into reporting tools.

Evidence QA v1 packages that preparation into a buyer-operated flow. Define your criteria and allowed labels. Select the exact speaker to review. The kit reads completed Zoom transcripts, asks OpenAI for a draft, checks the quoted text against the cited turns and permitted speakers, calculates points in code, and publishes a Google Doc report and Google Sheet row.

You receive the complete source kit, not just a prompt. Try the fictional support example offline before connecting any account. Inspect the implementation and run its tests. Customize the rubric in JSON and schedule the flow through the included inactive n8n workflow.

Why the checks matter

• Exact Unicode excerpts are verified against their cited turns. Subject criteria require the evaluated speaker's own words.
• Explicit conversation criteria can cite another speaker, such as a customer's confirmation, without attributing that outcome solely to the employee.
• Points come from your rubric. Score and coverage are shown separately, so missing evidence does not disappear into a flattering total.
• Human-review requests and unsupported evidence leave criteria unscored.
• The assessment is saved before external publication. A failed output resumes from the saved result rather than repeating the model after that checkpoint.
• Report and row identities support reconciliation after an uncertain write, with one local writer per state directory.

This prepares auditable coaching and conversation QA drafts. A checked quote proves that text exists in the transcript; it does not prove the model's interpretation is correct. Calibrate your rubric on reviewed examples and keep a person responsible for consequential decisions. This product does not decide employment eligibility, pay, promotion, or discipline.

Included

• Root setup skill and full Node.js implementation source.
• Zoom, OpenAI Responses, Google Drive/Docs, and Google Sheets adapters.
• Configurable rubric and fictional support transcript/assessment examples.
• Offline demo with HTML report, checked assessment JSON, and draft CSV.
• CLI scan and authenticated local HTTP engine.
• Inactive hourly n8n schedule export.
• Local contract and synthetic HTTP adapter tests.
• Local synthetic benchmark command.
• Setup, rubric, privacy, operating/recovery, evidence, and roadmap documentation.
• Editable logo, workflow screenshots, captioned process animation and an offline interactive business/process map.

See the process before connecting accounts

The public visual guide shows the implemented v1 flow, its evidence checks and publication recovery: workflow screenshots and animation. The illustrations use fictional data; they do not depict a shipped reviewer application or a verified live provider run. The included controllable preview starts paused and offers a static alternative.

Before purchase

You need Node.js 22+ and technical access to configure the flow. The offline demo needs no account or key. Live automation requires eligible Zoom transcripts/API access, an OpenAI API key and compatible model, Google OAuth setup with a refresh token, an existing draft Sheet tab, and a report folder accessible to your OAuth application. An existing Drive folder's ownership alone does not prove drive.file app access. OAuth consent and destination authorization are manual setup steps.

The purchase excludes provider fees, hosting, custom deployment, account administration, custom rubrics, and semantic calibration. Contract and synthetic tests establish local behavior. Actual n8n import/execution and live provider operation were not verified for this release; smoke-test them in your own environment. There is no universal accuracy, time-saving, or per-review cost guarantee.

V1 and future editions

V1 is the first complete standalone paid source edition. Updates published on this same Agensi listing are free to owners under the marketplace's terms. Future major editions, domain packs, or a hosted edition, if released, will be sold through separate paid listings. Buying those is optional; v1 remains usable. Roadmap features have no guaranteed dates and are not included until released under their stated terms.

Launch offer: $41.00 USD one-time (at least 30% off) until 2026-11-01T08:50:56.967000Z (01 November 2026 at 09:50:56 CET, Europe/Amsterdam). The regular one-time price is $59.00 USD afterward.