Jev

@Jev evaluate this draft against our 'brand_voice' and 'technical_accuracy' questions to see if it's ready for publication.

Evaluated the draft against 2 questions. **brand_voice** (Noul): Score 0.82 — tone aligns well with established voice guidelines, minor formality inconsistencies in paragraph 3. **technical_accuracy** (Score): 7/10 — core concepts correct, but the API endpoint example references a deprecated v1 path; recommend updating to v2. Both evaluations consumed inference tokens and were billed to your account.

This example shows parallel evaluation against named questions you've pre-configured in Jev. Each evaluation call is usage-priced, so running multiple questions simultaneously incurs cumulative inference costs. Useful for gating content workflows, but be mindful of token spend on high-volume checks.

@Jev list the models available in my account so I know which evaluation engines I can use for different question types.

Your Jev account has access to 4 models: **jev-standard** (alias: default), **jev-fast** (alias: quick), **jev-precise** (alias: thorough), and **jev-custom-finance** (custom fine-tune). The 'standard' model is used by default unless you specify an alias when configuring a question. Fast trades some accuracy for lower latency; precise is recommended for high-stakes evaluations.

This read-only call helps you understand which evaluation models are available before you configure questions or run evaluations. Model selection affects both accuracy and cost, so knowing your options upfront is critical for budgeting and quality tuning.

@Jev evaluate our latest feature spec against the 'user_clarity' question, then summarize whether we need another revision before sharing with stakeholders.

Evaluated the feature spec against **user_clarity** (Choice): Selected option 'Needs Minor Edits' with 78% confidence. The spec clearly defines user flows and success metrics, but the error-handling section uses internal jargon that non-technical stakeholders may not parse. **Recommendation**: Revise the error-handling paragraph to use plain language, then the spec should be stakeholder-ready. This evaluation consumed tokens and was billed.

This synthesis example pairs Jev's structured evaluation (a Choice question returning a categorical answer) with the AI's reasoning to produce an actionable recommendation. It demonstrates how Jev fits into decision workflows, though you'll need to pre-define the 'user_clarity' question and its answer choices in your Jev account first.

When parallel evaluation beats spreadsheet voting for design critiques

A 6-person design team runs weekly critiques where they score mockups against brand guidelines, accessibility standards, and user-test readiness. Before Switchy, they passed around a Google Sheet and manually tallied scores—messy when someone forgets to fill their row. With the Jev MCP, the team drops mockup links into a shared chat thread, then runs Evaluate State against three named Score questions (brand alignment 1-10, WCAG compliance yes/no, test-ready yes/no). The MCP returns all three scores in one call, usage-priced per evaluation. This works when your criteria are stable enough to predefine as Jev questions and your team reviews fewer than 50 assets a week—beyond that, the per-call pricing adds up faster than a Typeform Pro seat. If you're scoring the same dimensions repeatedly and want the tally inside your chat context, Jev keeps the workflow in one place.

Parallel sentiment scoring for high-volume ticket queues

A 3-person support team at a SaaS startup triages 200 inbound tickets a day, tagging each for urgency and customer sentiment before routing to engineering or success. They used to skim subject lines and guess—now they paste ticket text into Switchy and call Evaluate State with two Choice questions (urgency: low/medium/high, sentiment: frustrated/neutral/happy). The MCP runs both evaluations in parallel and returns structured answers in seconds, so the team can batch-tag 20 tickets in a standup. The trade-off: Jev charges per evaluation, so if your ticket volume spikes past 500/day, you'll want to compare the monthly cost against a dedicated sentiment API like MonkeyLearn. For teams under that threshold who already live in Switchy for other workflows, Jev keeps triage fast and contextual without leaving the workspace.

When batch evaluation beats manual rubric scoring for content refresh

A 4-person content team audits their blog every quarter, scoring 80 posts against SEO health, brand voice consistency, and factual accuracy. They used to split a Notion database and score posts one-by-one over three days—now they export post URLs to Switchy, loop through them with Evaluate State, and run three named Score questions per post (SEO 1-10, voice 1-10, accuracy yes/no). The MCP returns all three scores in one call per post, so the full audit finishes in an afternoon. This makes sense when your rubric is stable and your corpus is under 200 items per quarter—above that, the cumulative inference cost rivals hiring a junior editor for a week. If your team already has the evaluation criteria defined as Jev questions and you want the audit log inside your shared workspace, this MCP turns a multi-day slog into a single-session sprint.