Delego: Autonomous Team Planning with Digital Twins

Overview

We won 1st place at the AgentVerse Hackathon by UCL Artificial Intelligence Society, sponsored by Anthropic, Amazon Web Services (AWS), Cisco, Entrepreneurs First, and more!

In just 24 hours, we built Delego, a system that tackles a problem every team faces. Even with advanced tools, project managers spend hours in meetings assigning tasks and tracking progress. We wanted to make planning fully autonomous.

The Problem

Project managers and team leads spend countless hours in:

Task assignment meetings
Progress tracking sessions
Resource allocation discussions
Coordination overhead

This friction slows down teams and limits their ability to scale effectively.

What Delego Does

Delego automates team planning using digital twins that debate, align, and assign tasks intelligently.

No meetings required for task assignment
No micromanagement needed
Frictionless scaling across teams

Key Technical Highlights

Multi-Agent Backend with RAG Modeling

Deployed the system as a multi-agent backend with RAG modeling and CAIPE, enabling scalable, autonomous collaboration across teams.

Agentic Digital Twin Data Pipeline

Built an agentic digital twin data pipeline with Node and Redis to model each team member's:

Skills and expertise
Availability windows
Working preferences
Historical performance

Fine-Tuned Language Model

Fine-tuned GPT-4.1 Nano with digital twin data for realistic decision patterns that match how real team members would respond.

Modular Workflow Agents

Designed modular workflow agents and integrated MCP servers (e.g., GitHub MCP) for seamless task delegation and project management integration.

Architecture

                    +------------------+
                    |   Task Input     |
                    +--------+---------+
                             |
                    +--------v---------+
                    |  Digital Twins   |
                    |  (Debate Phase)  |
                    +--------+---------+
                             |
              +--------------+--------------+
              |              |              |
      +-------v------+ +-----v------+ +-----v------+
      | Twin Agent 1 | | Twin Agent 2| | Twin Agent N|
      | (Skills/Avail)| | (Skills/Avail)| | (Skills/Avail)|
      +-------+------+ +-----+------+ +-----+------+
              |              |              |
              +--------------+--------------+
                             |
                    +--------v---------+
                    |  Consensus &     |
                    |  Assignment      |
                    +--------+---------+
                             |
                    +--------v---------+
                    |  MCP Integration |
                    |  (GitHub, etc.)  |
                    +------------------+

Impact

This marks my 4th consecutive hackathon win and demonstrates the power of multi-agent systems for real-world productivity challenges.

Team

Huge thanks to the amazing team that made this win possible:

Salman Chishti
Katie Lam
Joe Clinton
Ryan Lin
Suleiman Khan

And appreciation to the UCL Artificial Intelligence Society's committees and volunteers for organizing such a well-run event!

Tech Stack

Backend: Node.js, Redis
AI/ML: GPT-4.1 Nano (fine-tuned), RAG, CAIPE
Integration: MCP Servers (GitHub MCP)
Architecture: Multi-agent system with digital twins

Delego