# Pipelex Projects at AI Tinkerers

> Canonical HTML: https://aitinkerers.org/technologies/pipelex
> Markdown URL: https://aitinkerers.org/technologies/pipelex.md
> Technology record last updated: 2026-04-04T19:51:29Z
> Generated: 2026-09-22T11:45:58Z

Pipelex is the open-source language for building and running deterministic, repeatable AI workflows: think Dockerfile for multi-step LLM pipelines.

Pipelex delivers production-ready AI workflows by replacing complex, unpredictable LLM prompts with structured, manageable steps (pipes). This open-source language uses human-readable `.plx` files and Concept-Driven Design (semantic typing) to ensure consistency and logical flow. It solves the reliability problem: developers can now test, validate, and deploy complex agentic flows, like invoice extraction or contract analysis, achieving 99% accuracy and deterministic results at scale.

- Official technology site: https://pipelex.com
- Public AI Tinkerers demos and talks: 1
- Result page: 1 of 1

## Recent Public Talks and Demos

### [Agent Builds Production AI Workflow in 4 Minutes](https://paris.aitinkerers.org/talks/rsvp__BV1raa1xGM)

I'll demonstrate an agent transforming natural language requirements into a production-ready AI workflow using our open-source declarative language for AI automation. We could call this Vibe-Coding a workflow or Vibe-Flowing. **The Live Demo** - Start with a business requirement such as expense report processing: *"Extract employee info and line items, process them one by one, validate against company rules, flag anomalies and write rejection email"* - Watch the agent design a multi-step pipeline with specialized pipes for each transformation - See it validate the workflow statically, then with a dry run on mock data, then with synthetic test cases - Deploy the final pipeline that handles document variations while guaranteeing structured outputs **What I'll Show Under the Hood** - The declarative TOML syntax that captures business intent, not API calls - How each pipe declares its conceptual contract: `PurchaseDocument → Buyer` instead of `string → JSON` - The agent's reasoning as it decomposes the mega-task into specialized steps - How it selects different models for different pipes: OCR, Vision, LLM, Image generation - The testing loop where the agent validates outputs and adjusts prompts **The Technical Architecture** - **Declarative abstraction layer**: Define what transformations you want, not how to implement them - **Deterministic structure with adaptive execution**: Generate validated Pydantic BaseModels at every run, while LLMs handle content variation - **Compositional design**: Pipes flow sequentially, batch in parallel, branch conditionally - **Agent-native format**: workflow definition that reads like documentation, perfect for LLM generation and refinement

- Event context: Epic Demo night 0.2.1 - Sep 18th, 2025 — 2025-09-18 — Paris
- Public talk page: https://paris.aitinkerers.org/talks/rsvp__BV1raa1xGM

## Related Technologies

- [instructor](https://aitinkerers.org/technologies/instructor) ([Markdown](https://aitinkerers.org/technologies/instructor.md)) — 3 public demos
- [Pydantic](https://aitinkerers.org/technologies/pydantic) ([Markdown](https://aitinkerers.org/technologies/pydantic.md)) — 30 public demos
- [Python](https://aitinkerers.org/technologies/python) ([Markdown](https://aitinkerers.org/technologies/python.md)) — 662 public demos
- [Visual Studio Code](https://aitinkerers.org/technologies/visual-studio-code) ([Markdown](https://aitinkerers.org/technologies/visual-studio-code.md)) — 10 public demos