AIGILE — Make Enterprises Agile Again

A project for the Yale x Google AI Hackathon.

AI-powered expertise graph that turns weeks of "organizational archaeology" into seconds.

Large enterprises are slow because nobody knows who knows what. Launching any cross-functional initiative — a new product, a sustainability program, a compliance response — starts with weeks of emailing department heads, scheduling discovery calls, only to be referred to someone else. Companies pay top consulting firms $100K+ for projects where the first weeks are spent on exactly this: responsibility and capability mapping.

Aigile cuts this from weeks to seconds.

How It Works

A user describes an initiative they want to kick off, and the system:

1. Finds the right people — searches an expertise graph built from employee work summaries to identify who has relevant knowledge, with evidence of why
2. Generates a kickoff plan — recommends project phases, assignments, timeline, and identifies knowledge gaps
3. Drafts personalized outreach — writes ready-to-send messages to each stakeholder, adapted to their role and communication style

Architecture

         ┌──────────────────────┐
         │   AIGILE ROOT AGENT  │  (Orchestrator)
         │   Text in → Text out │
         │   (Voice = stretch)  │
         └──────────┬───────────┘
                    │
            ┌───────┴───────┐
            ▼               ▼
      ┌──────────┐    ┌──────────┐
      │  MAPPER  │    │ BRIEFER  │
      │  Agent   │    │  Agent   │
      │          │    │          │
      │ Queries  │    │ Generates│
      │ expertise│    │ plan +   │
      │ graph    │    │ outreach │
      └──────────┘    └──────────┘

• Root Agent — takes the user's initiative description, orchestrates the pipeline, returns combined results
• Mapper Agent — extracts keywords → queries Meilisearch via MCP → ranks stakeholders with evidence chains
• Briefer Agent — generates a project kickoff plan (phases, timeline, assignments), identifies knowledge gaps, drafts personalized outreach per stakeholder

Privacy-First Design

In production, lightweight local AI agents on each employee's machine summarize their work topics daily. No sensitive data (emails, documents, code) leaves their machine — only topic summaries (e.g., "Julia worked on: bio-plastics sourcing, CSRD Scope 3, supplier audits"). These summaries are indexed centrally and made queryable.

For the hackathon demo, we simulate this with pre-generated employee profiles for a fictional 5,000-person German automotive supplier.

Tech Stack

Component	Technology
Agent Framework	Google ADK (Agent Development Kit) for Python
LLM	Gemini 2.0 Flash / Pro via Vertex AI
Search Engine	Meilisearch (local Docker instance)
Agent-to-Search	MCP (Model Context Protocol) server wrapping Meilisearch
Backend	Python (FastAPI / ADK built-in server)
Frontend	React + Tailwind CSS
Deployment	Google Cloud Run
Voice (stretch)	Gemini Live API via ADK Streaming

Getting Started

Prerequisites

• Python 3.10+
• Node.js 18+
• Docker (for Meilisearch)

Backend

cd backend
pip install -r requirements.txt
# Add your Google API key
echo "GOOGLE_API_KEY=your-key-here" > .env
python server.py

Frontend

cd frontend
npm install
npm run dev

Demo

Example query: "I need to kick off a CSRD Scope 3 compliance initiative for our supplier base, targeting the 2026 reporting deadline."

The system finds 5–6 key people across 4 departments, generates a phased project plan, identifies knowledge gaps, and drafts outreach messages — all in under 30 seconds.

Business Model

• SaaS subscription per enterprise ($50K–200K/year depending on org size)
• Vertical wedge: regulatory/compliance initiatives in automotive and manufacturing (CSRD, CSDDD, EU AI Act)
• Expand to all cross-functional initiative types (product launches, M&A integration, restructuring)
• Moat: compounding playbook library of how specific regulation × industry combinations get handled operationally

Context

Built at the Yale Build with AI Hackathon × Google Cloud Labs (April 2026) for "The Live Agent" track — real-time, voice-and-vision enabled agents using Google Cloud, Gemini, and ADK.

Origin Story

"At BMW, I was tasked with launching a sustainability initiative for interior materials. It took me 3 weeks and 5 departments just to figure out what our own company knew. Companies pay McKinsey six figures for exactly this."

---

See AIGILE_MASTER_CONTEXT.md for full project context and pitch details.