Self-Programming AI: A Meta-Agent Approach to Visual Data Extraction

The talk demonstrates a meta‑agent that employs Vision Language Models and DSPy to program AI agents for extracting data from Bureau of Labor Statistics images.

Overview

Inspired by the AI Tinkerers Paper Club’s presentation on “Automated Design of Agentic Systems”, I want to explore a system where a meta-agent designs and programs other AI agents to extract data from Bureau of Labor Statistics images. This talk demonstrates how Vision Language Models (VLMs) and the DSPy framework enable this advanced automation, offering a glimpse into self-adapting AI for complex document understanding tasks.

Tech stack