Sedol Maestro: AI-Human Collaborative 'Baduck' Coding

Interactive AI-human collaborative coding platform using UiPath Maestro Case to science-calculate the single best "move" and self-heal build failures.

Comment

Inspiration

Our journey began with a question: **Can we bring the depth, strategy, and precision of a legendary Go (Baduck)

game into software development?** In 2016, Lee Sedol defeated AlphaGo with his legendary 78th move (the "Divine Move")—a decision so creative and complex that no computer could calculate or predict it.

Modern AI coding agents (like Claude Code, Cursor, or Gemini CLI) can generate massive amounts of code instantly,

but they frequently hallucinate or fall into catastrophic build loops on edge cases. Like AlphaGo, they lack human intuition at critical junctures.

**Sedol Maestro** was inspired by this exact concept: to map a complex software codebase onto a 19x19 Go board,

calculate the scientific & probabilistic win-rate of the current "architectural state," and orchestrate a flawless human-agent collaboration loop where humans can intervene with their own "78th move" to rescue agents and ensure enterprise-grade stability.

## What it does
**Sedol Maestro** is a unified AI-human collaborative workbench that fuses three unique systems:
1. **Perception (`app-factory`)**: Parses a repository's AST, module complexity, and test coverage, mapping the

codebase into a visual Go board state (board_state.json). 2. Strategy (baduck-coding UI): Simulates next best steps via a custom Monte Carlo Tree Search (MCTS). It displays the top candidate moves (Prompts, Refactoring targets) with AI-evaluated win rates directly on a beautiful Go Board interface. 3. Orchestration (UiPath Maestro Case): Serves as the control plane. It guides the project lifecycle through stages. When an agent enters an exception state (e.g., failing a build 3 consecutive times), UiPath suspends the flow and triggers a Human-in-the-Loop Task (Action Center). The human makes the ultimate "move," resuming the agent to flawless execution. 4. Governance (gido-board): Commits the successful decisions to a "Kifu" (Go game log) for community review and provides a humorous yet practical "Prayer Board" (祈禱 / 棋道) where developers request build blessings and receive AI-guided architecture audits.

## How we built it
We architected Sedol Maestro as an elegant **Vite-React-TypeScript Monorepo**:
- **Frontend**: A highly responsive, premium glassmorphic portal blending a 19x19 SVG Go Board, dynamic win-rate

meters, and a real-time Markdown prayer forum. - Backend (Express + Simple-Git): Analyzes codebase nodes, runs local build probes, and feeds state metrics to LLM APIs (Claude & Gemini). - Automation (UiPath Automation Cloud): Orchestrated using UiPath Maestro Case Management. We built custom activities that monitor git commits, intercept webpack/tsc compile outputs, handle error triggers, and publish Tasks to UiPath Action Center for real-time human micro-decisions.

## Challenges we ran into
Integrating real-time terminals, long-running agentic execution, and enterprise RPA was a major challenge.

Ensuring that the state of a fast-paced development terminal (where simple errors happen) maps smoothly into a long- term UiPath Case Instance without losing context required us to build a robust REST Bridge Adapter that acts as the source of truth between local coding agents and the UiPath Orchestrator Cloud.

## Accomplishments that we're proud of
We are incredibly proud of successfully bridging the gap between low-code enterprise tools (UiPath Action Center)

and highly advanced coding agents (Claude Code, Gemini CLI). Designing a fully unified mathematical model that maps real-world code AST structures and PageRank metrics directly onto a 19x19 Go Board coordinate system with real-time exception-handling is a massive achievement we are extremely proud of.

## What we learned
We learned that the future of software engineering is not fully autonomous agents, nor is it manual labor. The

sweet spot is Orchestration. By utilizing UiPath Maestro, we could easily build resilient, exception-heavy workflows that survive server crashes and API limits, allowing humans to govern AI with zero friction.

## What's next for Sedol Maestro: AI-Human Collaborative 'Baduck' Coding
We plan to integrate **UiPath Test Cloud** to run risk-weighted regression suites on every "move" in the

background, automatically suggesting defensive "moves" (safety stones) on the board before code reaches staging.

Built With

Share this project:

Updates