Twitxy
Saved
@pydantic's profile picture
View on X

Pydantic

This account is verified because it's an official organization on X. Learn more

@pydantic

The Pydantic Stack: Logfire, Monty, Validation, AI, Evals, and AI Gateway
London pydantic.dev Joined August 2022
177 Following 21K Followers 2.9K Posts

Latest posts 5

Pinned post

Pydantic @pydantic
One Jev request, three scores: a policy label, a rubric level, and the probability the reply asks for a password, all recorded in Logfire.

Modeled cost of 1M evaluated replies: $62, vs $4,542 with Braintrust before data charges.

Blogpost pydantic.io/n4PWy
pydantic.dev Cheap AI scoring with Jev and Pydantic Evals Build a Jev evaluator with categorical and numeric scores. Run it with Pydantic Evals and compare high-volume scoring costs in Logfire and Braintrust.
6 replies 6 reposts 53 likes 5.5K views View on X
Pydantic @pydantic
Elumynt runs 100+ daily agentic routines. Their Pydantic AI agent now writes each deterministic step as a Python program and runs it in Pydantic Monty. LLM spend on routines is down by over 80%, and the same inputs give the same output every day.

pydantic.io/elumynt
pydantic.dev How Elumynt cut LLM costs by over 80% with Pydantic Monty | Pydantic Elumynt's agents write Python for each deterministic step and run it in Pydantic Monty, cutting LLM spend on 100+ daily routines by over 80%.
2 replies 1 reposts 21 likes 1.4K views View on X
Pydantic @pydantic
Security teams have gotten back ~332,000 analyst hours from @SentinelOne’s agents, at 99.7% agreement with human experts.

They get scored one capability at a time, each with its own rubric and judges scoped to it, gated in CI/CD by Pydantic Evals.
pydantic.io/sentinelone-x
pydantic.dev How SentinelOne evaluates the AI agents behind a 99.7% agreement rate | Pydantic SentinelOne scores its security agents with Pydantic Evals in CI/CD and self-hosted Pydantic Logfire, shipping new AI features 30% faster.
3 replies 0 reposts 20 likes 1.1K views View on X