Small decisions. Interesting possibilities.Submit contentSubmit

How it uses Jev

The query analyzer sends the user's log query and known service names to Jev. Choice questions classify the requested answer path, severity, and service; a Noul guard checks whether the query really names a service. Confidence gates control whether the analyzer applies intent and severity, and both the guard and service confidence must pass before it adds a service filter. Time-range parsing remains in local code.

What Jev decides

Illustrative log-query classification requestExample answers · not a recorded Jev response · Source ↗
Question 1 · intent
YOUR APP
INSTRUCTION

Classify the kind of answer the query asks for.

STATE

The state includes the user's query and up to 254 known service names. The service question is included only when that list is nonempty.

JEV · CHOICE
  1. causal
  2. trace
  3. summary
  4. search
  5. unclear
Question 2 · level
YOUR APP
INSTRUCTION

Choose the requested log severity, or any when the query does not restrict severity.

STATE

The state includes the user's query and up to 254 known service names. The service question is included only when that list is nonempty.

JEV · CHOICE
  1. Error
  2. Warn
  3. Info
  4. Debug
  5. any
Question 3 · service named
YOUR APP
INSTRUCTION

Check whether the query names or clearly refers to a known service.

STATE

The state includes the user's query and up to 254 known service names. The service question is included only when that list is nonempty.

JEV · NOUL
YesNo

App workflow

  1. Parse deterministic fields

    Rules code extracts a time range and produces a fallback classification before Jev is called.

  2. Ask typed classification questions

    Jev receives the query and known service list, then returns intent, severity, and conditional service decisions in one request.

  3. Apply confidence gates

    The analyzer applies intent and severity above their configured floors. It adds a service filter only when the Noul service guard and Choice confidence both pass.

  4. Fall back on failure

    If the Jev call fails, the analyzer returns the rules-based result.

Why it is interesting

It separates intent and severity decisions from service filtering, then applies a stricter gate to service filters because a wrong filter can hide relevant logs. It also preserves a rules-based fallback when the Jev call fails.