Overview & Result
I used Jev to classify 1,018 AI research papers. The result: $0.08 total cost and 256ms median end-to-end latency per paper. The pipeline was: 1. Summarize each paper with DeepSeek V4 Flash 2. Send the title + summary + 24 possible topics to Jev 3. Use Jev to classify each paper 4. Visualize everything on http://1kpapers.com The summaries cost $3.99 on @togethercompute. The classifications cost $0.08 on @typesafeai. So for just over $4 of inference, I ended up with a pretty useful way to explore the top AI research papers from the past year. I think this is where things are heading: different models for different parts of the workflow, instead of using one model for everything. I’m running evals on the Jev classifications before replacing the current ones, but the site is already live: http://1kpapers.com
How Jev fits in the loop
- Extract observation from agent environment.
- Jev evaluates state and outputs typed choice or probability.
- Agent runtime carries out selected action.
How to reproduce
- Inspect author post and reproduction notes.
- Deploy agent loop and bind Jev decision endpoints.
- Run test inputs and observe latency and accuracy.
Why this build matters
Demonstrates practical production-level utility of Jev inside specialized agent workflows.
Limitations
- Community observed build; metrics are author-reported.
- Requires third-party dependencies as described in source.