Jev for data teams: what it can do and what it costs

Jev, TypeSafe AI's decision model, answers a question about text with a value your software can use: one option from a list, a score, or a yes/no probability. That suits the small, repeated decisions in the middle of this work.

These are the jobs on this site that fit data teams best. Each has its own page with the questions to ask, the pitfalls and a worked example.

• Decide which records to merge, which to keep apart and which a person should check. For each candidate pair, one Score question with three levels (different, possibly related, same) decides whether to leave the records apart, ask a person or merge them, and a few yes/no questions in the same request show which fields disagree.
• Pick one of dozens of categories, and fall back to a broader one when unsure. In TypeSafe's cookbook, 60 company annual reports were classified into 75 industry groups with one Choice question each.
• Find the right value in messy text without letting a model make one up. For values such as emails, phone numbers and amounts, a pattern match finds the candidates and Jev picks the one the question asks for, so it cannot invent a value.

• Jev for matching duplicate records
• Jev for classifying documents into many categories
• Jev for extracting dates, emails and amounts

Jev lists $0.042 per million input tokens with output free, so the cost depends on how much text you send and how many questions you ask about it. At the typical volumes the use-case pages assume, the calculator estimates:

• Pairs (matching duplicate records): about 1.2¢ per 1,000, at 4 questions and about 120 tokens each.
• Documents (classifying documents into many categories): about 11¢ per 1,000, at 1 question and about 1500 tokens each.
• Documents (extracting dates, emails and amounts): about 2.9¢ per 1,000, at 3 questions and about 500 tokens each.

• Cost calculator
• Jev pricing and real runs

A plain way to begin, in order:

• Pick one cleanup that is blocking a report, such as deciding whether two records are the same thing.
• Turn it into a Score or Choice question with the levels or categories your team already uses.
• Use the confidence value to split the output into answers you trust, answers a person should check, and a broader fallback.
• Write the checked cases back as labelled data, so the next run starts better.

• How to use Jev

For extraction, let Jev choose among candidates your code found rather than ask it to produce a value. That is how TypeSafe's cookbooks keep it from inventing one.

From the use-case pages, Jev is not the right choice when:

• Records share a reliable ID. Match on the ID in code.
• Every document must get a precise label with no review. Plan for the unsure half.
• The document has a fixed layout. A parser is cheaper and exact.

• What Jev can't do

Data cleaning is the place where the confidence value earns its keep. Being able to say 'half of these are safe to merge, half need a person' is worth more than a slightly higher accuracy number.

The figures above are list-price estimates, not measured bills, and the accuracy claims on the linked pages are the published runs' own. Test Jev on a labelled sample of your own data before you rely on it.

• What is a decision model?