evaris

Build reliable ai systems

[roshan @pucchkaa]

Ran some evals on Jev vs GLM-4.6 on Healthbench and got some crazy results.
jev in terms of -
accuracy : statistically same
latency : 100x faster
cost: 100x cheaper
parse failures: 0 on both models

@typesafeai @CompleteSkeptic

[roshan @pucchkaa]

If you haven't setup evals for your product or are looking to explore evals, you can try out @evaris_ai . It helps you setup evals using inspect-ai and evaris platform and takes hardly 1 min.

You can quickly setup evals in your CI using our Agent skill.

Its simple. Try it.

[evaris @evaris_ai]

zero config. just init() and you're tracing.

MIT licensed → https://t.co/AXIIRBT7Zf

[evaris @evaris_ai]

🚀 introducing etrace — open-source AI agent tracing library

one primitive: trace
auto-instruments LLM calls
tracks costs across 1700+ models
16 semantic span kinds built for AI workflows

-

📊 etrace studio — visualize, inspect, and debug every trace locally

[evaris @evaris_ai]

hello world!