Sisyphus Intelligence Mapping
aipythond2geminivisualizationmulti-agent
January 2026 Built

Sisyphus Intelligence Mapping

Automated pipeline that converts agentic session traces into cyberpunk HUD infographics. D2 diagram generation + Gemini orchestration + Python rendering. Proves 12-minute agentic labor vs 6-hour manual work — visually.

Built with

Python
Python
D2
D2
Gemini
Gemini
Multi-Agent
Multi-Agent

The Problem

Agentic AI workflows are black boxes. When Claude or Gemini completes a complex multi-step task, the output is visible but the process isn’t. Non-technical stakeholders can’t understand what happened. Technical stakeholders can’t audit it easily. The intelligence is invisible.

The Solution

Sisyphus is an automated visualization pipeline that takes raw Claude Code session logs and transforms them into structured cyberpunk HUD infographics — showing the full decision tree, tool calls, correction loops, and outcome verification in a single shareable image.

How It Works

01
Session Ingest
Parse Claude Code JSON — tool calls, reasoning chains, outcomes
02
Schema Extract
Gemini Flash identifies stages, decision nodes, branching logic
03
D2 Generate
Auto-generate D2 diagram code for the full workflow
04
Render
D2 CLI → SVG → Python styled PNG infographic

The “FINAL WHOA” Blueprint

The V4 output (sisyphus_FINAL_WHOA.png) shows a 6-stage pipeline schematic:

  • Input Layer — Session trace ingestion
  • Truth Engine — Self-correction and verification loop
  • Intelligence Map — Staged decision visualization
  • Efficiency Layer — Time comparison (agentic vs. manual)
  • Output — Final rendered infographic

Key Outcome

12 minutes of agentic work visualized as equivalent to 6 hours of manual work — with zero manual code written during the session that produced the output.

The pipeline itself was built agentic-first: Claude wrote the D2 schema, generated the Python rendering code, and debugged the SVG pipeline in a single session.

← All Projects