Persuader
FILECONORFYI · 2025
PERSUADERFILED 2025
READ1 MIN READ
WORDS68
SECTIONS4
AI—
012 KM
Historical project. Prompts your LLM until it outputs schema-conforming JSON.
§ 01
Historical project
An early framework for processing collections through repeated LLM sessions, using Zod validation and error feedback to retry responses. Preserved as an earlier approach to agent orchestration, rather than current setup guidance. The underlying ideas of validation, retries and session continuity remain useful.
§ 02
Overview
Retries LLM calls with full error context until Zod schema validation passes. Persists sessions to avoid re-sending initial context on every retry — useful when front-loading system prompts is expensive.
§ 03
Highlights
- Zod-based schema validation on every response
- N retries with structured error context fed back to the LLM
- Session persistence to minimise re-sent context
- Works with any LLM that returns JSON
§ 04