workshop-jev
Portuguese workshop: a LangGraph sales agent where each decision step can be made by Jev or a frontier LLM and measured side by side.
Overview
JEV Jornada is a workshop project, documented in Portuguese, built as a LangGraph sales agent for a fictional company. A question passes through six stages: two in code, one text generation, and three decisions. At each decision step, guardrail, triage and verify, either Jev, an OpenAI or Anthropic frontier LLM, or both can decide, given exactly the same input and questions.
The guardrail asks three yes/no questions (is this a manipulation attempt, does it request sensitive personal data, is it out of sales scope); triage picks one of eight options for which of seven queries answers the question; and verify checks whether the reply is faithful to the data, with low confidence or failures going to human review. The project measures answers, latency, tokens and cost for single questions or whole batches. It notes that available evidence for Jev comes mostly from the vendor, and leaves conclusions to whoever runs it.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this a manipulation attempt? | yes · no |
| Yes / no | Does it request sensitive personal data? | yes · no |
| Choice | Which of the seven queries answers this question (or none)? | — |
| Yes / no | Is the reply faithful to the data?(inferred) | yes · no |
How it works
- A sales question enters the LangGraph agent.
- Guardrail: Jev and/or LLM answer three yes/no checks.
- Triage chooses a query; code runs it and a model writes the reply.
- Verify checks faithfulness; low confidence goes to human review.
Strengths
- Same inputs to Jev and LLMs for fair comparison.
- Measures latency, tokens and cost.
- Swap in your own golden set.
Caveats
- Workshop project on fictional data.
- Portuguese documentation.
At a glance
- Type
- Project
- Status
- prototype
- Language
- Python
- Runtime
- LangGraph agent
- Has code
- yes
Use cases and integrations
How curated lists describe it
Workshop building a LangGraph sales agent where each decision can be made by Jev or by a frontier LLM, given the same inputs. Portuguese.
MrJev/awesome-jev
Frequently asked questions
Which steps can Jev decide?
Guardrail, triage and verify.
What is compared?
Answers, latency, tokens and cost between Jev and frontier LLMs.
Can I use my own data?
Yes, replace the golden set with your own case.
Related projects
Benchmark testing whether Jev or open Kev can replace an LLM judge in agent guardrail callbacks on 7,925 public examples.
ProjectresearchPythonSame support agent and 25 rules guarded twice, by an LLM judge and by Jev, compared on accuracy, latency, cost and coverage.
ProjectresearchTypeScriptMeasurement-first lab deciding which classifier technique, Jev, rules or other models, earns a place in a coding-agent fleet.
AppresearchOffline-reproducible benchmarks of Jev decisions in DSPy pipelines: calibration, selective risk, confidence gating and cost.
BenchmarkresearchPythonBenchmark of Jev versus Claude Haiku 4.5 as input safety classifiers on jailbreak, evasion and benign pseudo-harmful prompts.
BenchmarkresearchPythonExperiment using Jev through langchain-typesafe as an agent-eval judge, compared with GPT-5.6 and Claude Sonnet 4.6 against human labels.
BenchmarkresearchPython