Overview & result
An agent skill and Python workflow for lead scoring with Clay. The Clay CLI retrieves people who match fixed filters such as job title, company size and location; Jev then judges whether each person really fits the intended role. The included use case targets current operating founders and filters out founding investors, founding employees, former founders and Founder's Office staff before enrichment.
From the original post
Agent Skill and Python workflow for Clay lead scoring, B2B prospect qualification, and people-search ranking with TypeSafe JEV.
How Jev fits in the loop
- The Clay CLI retrieves people who match exact filters such as job title, company size, industry and location.
- For each person, Jev answers Choice and Noul questions about whether they really fit the target role.
- Threshold flags turn those probabilities into a qualified or unqualified decision before enrichment.
- To target another audience you change the Clay query and replace the founder-specific rubric.
How to reproduce
- Inspect the original showcase and source material at https://github.com/promptgtm-shared/clay-jev-people-ranker
- Verify the bounded prompt and input candidate schema configured for Jev
- Benchmark decision latency and classification accuracy against baseline models
Why this build matters
It splits the work cleanly: exact filters stay in Clay, and Jev is used only for the semantic call that filters cannot make, such as telling an operating founder from a founding investor.
Limitations
- A Jev probability threshold is not measured accuracy; the README says precision and recall must be measured on labelled examples.
- Jev only evaluates the evidence included in its state, and missing Clay classifications stay unknown.
- The founder rubric does not rank commercial attractiveness, buying intent, growth or funding.