HN Showcase
Back to browse
SafeParse – schema validation and retries for AI pipelines

SafeParse – schema validation and retries for AI pipelines

by benedictmendoza·Mar 8, 2026·1 point·2 comments
Visit ProjectView on HN

AI Analysis

●MidShip ItSolve My Problem

Schema validation + retry for AI pipelines, but already solved by Pydantic, Marvin, and structured outputs.

Strengths
  • •Real problem identified: JSON validation failures in production LLMs even with structured outputs is genuine friction for pipeline builders.
  • •Multi-model fallback (OpenAI → Anthropic → Google) and full request replay for debugging adds useful observability layer beyond raw validation.
Weaknesses
  • •Core functionality (schema validation + retry logic) is now table stakes in every LLM framework: Pydantic v2, OpenAI structured outputs, Anthropic Prompt Caching, Marvin, instructor, all do this natively.
  • •Landing page is waitlist-only with no live demo, no code, no pricing, no proof of concept—'testing whether people want this' is not a product, it's a hypothesis. Mid tier fits the embryonic stage.
Target Audience

Automation engineers, AI pipeline builders, teams running production LLM workflows

Similar To

Pydantic · OpenAI structured outputs · Marvin

Post Description

Hi HN,

I’ve been building a few automation pipelines that rely on LLMs, and I kept running into the same issue.

Everything works fine during testing, then breaks in production.

Typical things I saw:

– the JSON shape changes slightly – required fields come back empty – the model times out or rate-limits – downstream automations silently fail

Even with JSON mode or structured outputs, this still happens occasionally.

After dealing with this a few times I started building a small service that sits between the model and the rest of the pipeline.

The idea is simple:

define a schema send text to a webhook get validated JSON back

If the response doesn’t match the schema, SafeParse retries with context or falls back to another model. Every request is logged so failures can be replayed and debugged.

The goal is basically to make LLM-powered pipelines behave more like production infrastructure instead of fragile prompt wrappers.

I just put up a landing page + demo while I test whether this is a real problem for other people building AI pipelines.

https://safeparse.com

Curious if others here have run into similar reliability issues with LLM-based workflows.

Similar Projects