Skip to content
Krishang ZinzuwadiaStart a conversation

B3 Hack · 2nd place

Labyrinth

A 27-agent system that turns a product brief into a deployed application.

role-specific agents
27
deployment pipeline
Automated
01

Context

The challenge

Keeping 27 agents on one delivery path.

Turning a short product request into running software requires many connected decisions: architecture, implementation, review, integration, and deployment. Labyrinth explored whether those responsibilities could be divided across specialized agents without losing the end-to-end delivery thread.

Krishang's role

  • Built the autonomous multi-agent studio and its cross-stack delivery flow.
  • Orchestrated 27 specialized agents with LangGraph.
  • Engineered Multiverse View for parallel architecture comparison and connected the result to automated deployment.
02

System

Architecture

From input to outcome.

  1. 01

    Brief

    A natural-language product request establishes the software the system needs to deliver.

  2. 02

    Specialize

    Twenty-seven role-specific agents divide the work under LangGraph orchestration.

  3. 03

    Compare

    Multiverse View keeps parallel architecture options visible for A/B comparison.

  4. 04

    Build

    Tauri, Bun, Next.js, and FastAPI form the application and orchestration stack.

  5. 05

    Deploy

    The delivery path ends in an automated cloud deployment pipeline.

03

Choices

Engineering decisions

What shaped the build.

01

Assign explicit agent roles

The system uses 27 specialized agents rather than asking one general agent to own the entire delivery process.

02

Preserve competing architectures

Multiverse View makes parallel approaches comparable before the system commits to a delivery path.

03

Treat deployment as part of delivery

The workflow includes a zero-touch cloud deployment pipeline instead of stopping at generated source code.

04

Result

Recorded outcomes

What the project delivered.

  • Placed second at B3 Hack.
  • Coordinated 27 specialized agents in one software-delivery system.
  • Connected product generation to an automated deployment pipeline.

Working stack

  • LangGraph
  • Tauri 2.0
  • Bun
  • Next.js
  • FastAPI
  • Python

Evidence boundary

What this account is based on.

The scope and results here are limited to the project record in Krishang's résumé; no public Labyrinth repository is linked yet.