Skip to content
Architecture

CrewAI Example: Agentic ALU Optimization Demo

embodied-ai-architect/docs/plan/crew-ai/crew-ai-example-claude.md

Great question. Let me build a complete, runnable example that exposes both the power and the pitfalls of agentic workflows.

The Example: Optimizing a Simple ALU

Section titled “The Example: Optimizing a Simple ALU”

I’ll use a small Arithmetic Logic Unit because:

  • It’s small enough to fit in LLM context
  • It has real optimization opportunities (carry chains, resource sharing)
  • It can be evaluated with actual open-source EDA tools
  • It will expose where LLMs hallucinate vs. where they genuinely help

Let me create the complete working system:I see the pipeline failed early because of issues in the lint step. Let me fix the pipeline to be more robust and show the full demo:The demo is working. Let me make the simulation more realistic by having it actually detect the shifter bug:Let me fix the pattern matching to properly detect the bug:

Let me package everything into a single downloadable archive for you:Here are both formats:

To use:

Terminal window
# Extract
unzip soc_optimization_demo.zip
cd soc_optimization


# Install dependency
pip install pydantic


# Run
python demo.py

Contents:

  • designs/alu_baseline.v — Original ALU RTL
  • designs/alu_tb.v — Testbench
  • tools.py — EDA tool wrappers
  • pipeline.py — Agent workflow + mock LLM responses
  • trajectory_eval.py — Trajectory evaluation
  • demo.py — Main demonstration

The demo uses mock EDA tools by default. If you have Yosys and Icarus Verilog installed, it will automatically use real synthesis and simulation.