Applications
Jev Playground
<span class="tag">{provider === 'vercel' ? 'experimental_evaluate' : provider === 'typesafe' ? '/v1/systemone' : 'alpha.decisions'}</span>The link points at the commit we read, so the line number still holds.
These question sets are lifted from this project's source exactly as written.
categorychoiceMain category of this ticket
bug — Product failure or unable to usebilling — Charges, payments or refundsfeature — Feature requestspam — Spam or meaningless contentbug_severityscoreIf this is a bug, how severe is it
0 — cosmetic: minor visual issue1 — minor: has a workaround2 — major: core feature affected3 — blocking: completely unusablewants_refundtrue/falseIs the user asking for a refund?
frustrationscoreThe user's mood
0 — calm1 — annoyed2 — furiousimport { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state,
questions: {
category: { type: 'choice', instructions: 'Main category of this ticket',
criteria: { bug: 'Product failure or unable to use', billing: 'Charges, payments or refunds', feature: 'Feature request', spam: 'Spam or meaningless content' } },
bug_severity: { type: 'score', instructions: 'If this is a bug, how severe is it',
criteria: ['cosmetic: minor visual issue', 'minor: has a workaround', 'major: core feature affected', 'blocking: completely unusable'] },
wants_refund: { type: 'boolean', instructions: 'Is the user asking for a refund?' },
frustration: { type: 'score', instructions: 'The user\'s mood',
criteria: ['calm', 'annoyed', 'furious'] },
},
});finechoiceMost specific issue type
duplicate_charge — Charged more than once for one purchasedeclined_card — Payment declined at checkoutrefund_status — Asking where a refund isinvoice_request — Wants an invoice or receiptcoarsechoiceBroad department
payments — Anything about charges, cards or refundsaccount — Login, profile, settingsproduct — Using the product itselfimport { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state,
questions: {
fine: { type: 'choice', instructions: 'Most specific issue type',
criteria: { duplicate_charge: 'Charged more than once for one purchase', declined_card: 'Payment declined at checkout', refund_status: 'Asking where a refund is', invoice_request: 'Wants an invoice or receipt' } },
coarse: { type: 'choice', instructions: 'Broad department',
criteria: { payments: 'Anything about charges, cards or refunds', account: 'Login, profile, settings', product: 'Using the product itself' } },
},
});fitscoreHow well does this company match a mid-market logistics customer
0 — poor1 — partial2 — good3 — idealurgencyscoreHow soon do they intend to buy
0 — someday1 — this year2 — this quarter3 — this monthhas_budgettrue/falseDo they state that budget is approved or available?
decision_makertrue/falseDoes the writer appear to be a decision maker rather than a researcher?
import { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state,
questions: {
fit: { type: 'score', instructions: 'How well does this company match a mid-market logistics customer',
criteria: ['poor', 'partial', 'good', 'ideal'] },
urgency: { type: 'score', instructions: 'How soon do they intend to buy',
criteria: ['someday', 'this year', 'this quarter', 'this month'] },
has_budget: { type: 'boolean', instructions: 'Do they state that budget is approved or available?' },
decision_maker: { type: 'boolean', instructions: 'Does the writer appear to be a decision maker rather than a researcher?' },
},
});modelchoicePick the most suitable model tier for the latest user turn
cheap — Everyday Q&A, typo fixes, light rewrites, small talkstrong — Multi-step reasoning, writing or debugging code, long-document analysis, mathnone — Greeting or acknowledgement that needs no model at allimport { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state,
questions: {
model: { type: 'choice', instructions: 'Pick the most suitable model tier for the latest user turn',
criteria: { cheap: 'Everyday Q&A, typo fixes, light rewrites, small talk', strong: 'Multi-step reasoning, writing or debugging code, long-document analysis, math', none: 'Greeting or acknowledgement that needs no model at all' } },
},
});