ET Gen AI Hackathon 2026 – Autonomous Workflow Agent

Team: HealthStack103

Building scalable, end-to-end technology solutions across AI/ML, healthcare, and open-source infrastructure.

Team Details

Yakshith K D – Team Lead & ML Developer , Nithesh – Embedded AI Developer , Sagar K – Flutter Developer

• Agent solution : Agentic AI System for Autonomous Enterprise Workflows that executes multi-step tasks, self-heals failures, and maintains full auditability.

Problem Statement

Agentic AI for Autonomous Enterprise Workflows

Design a multi-agent system that:

• Owns complex enterprise workflows end-to-end
• Detects failures and self-recovers
• Minimizes human intervention
• Maintains auditable decision trails

---

Project Overview

This project evolves a Slack-based AI assistant into a fully autonomous workflow orchestration system.

Instead of just responding to messages, the system:

• Executes multi-step enterprise workflows
• Coordinates multiple specialized AI agents
• Detects failures and self-recovers automatically
• Logs every decision for auditability

---

Use Case (Demo Workflow)

Example: Meeting → Task → Execution Pipeline

1. User shares meeting discussion in Slack

2. AI extracts:

   • Key decisions
   • Tasks
   • Responsible owners

3. Automatically assigns tasks

4. Tracks progress continuously

5. Detects delays or inactivity

6. Escalates or reassigns tasks

All without manual follow-up.

---

Core Idea: Multi-Agent System

User → Orchestrator Agent → Specialized Agents → Execution → Monitoring → Self-Healing

---

Key Features

1. Multi-Agent Orchestration

• Central Orchestrator Agent manages workflows

• Specialized agents:

  • Task Extraction Agent
  • Decision Agent
  • Execution Agent
  • Monitoring Agent
  • Audit Agent

---

2. Autonomous Workflow Execution

• End-to-end task lifecycle management
• Minimal human input required
• Automatic task assignment and tracking

---

3. Self-Healing System

System detects:

• Delays
• Failures
• Missing actions

Automatically performs:

• Retries
• Reassignments
• Escalations

---

4. Audit Trail (Hackathon Key Requirement)

Every step is logged in a structured format:

[Time] → Agent → Action → Reason → Outcome

Example:

10:02 → Task Agent → Created task → Based on meeting notes
10:10 → Monitor → Delay detected → No update from owner
10:11 → Orchestrator → Reassigned → Backup user assigned

---

5. Tool Integration (via MCP)

• GitHub → Issues, PRs
• Notion → Docs, databases
• Slack → Communication

---

6. RAG + Memory System

• Retrieves past Slack conversations (RAG)
• Stores long-term user preferences (mem0)
• Uses context to make smarter decisions

---

Architecture

High-Level Architecture

┌──────────────┐
│   USER       │
└──────┬───────┘
       ▼
┌──────────────┐
│ Slack Layer  │
└──────┬───────┘
       ▼
┌────────────────────────────┐
│  ORCHESTRATOR AGENT        │
└──────┬────────────┬────────┘
       ▼            ▼
┌────────────┐  ┌────────────┐
│ Task Agent │  │ Decision   │
└────────────┘  └────────────┘
       ▼            ▼
┌────────────┐  ┌────────────┐
│ Execution  │  │ Monitoring │
└────────────┘  └────────────┘
       ▼            ▼
     Tools       Audit Logs

---

Workflow Lifecycle

1. Input received (Slack / API)
2. Context built using RAG + Memory
3. Orchestrator decides workflow plan
4. Specialized agents execute tasks
5. Monitoring agent tracks progress
6. Errors handled automatically
7. Audit logs stored

---

Impact Model

Time Savings

Task	Manual	AI System
Task extraction	30 min	2 min
Assignment	10 min	Instant
Follow-ups	Daily	Automated

Approximately 70% reduction in workflow time.

---

Cost Reduction

Assumption:

• Manager hourly cost: ₹1000
• 2 hours saved per day

Savings = ₹1000 × 2 × 22 days = ₹44,000/month per team

---

Efficiency Gains

• No missed tasks
• Faster execution
• Real-time monitoring
• Automatic recovery

---

Example Scenario

User Input:

"We need to fix login bug, update docs, and deploy by Friday"

AI System Output:

• Creates structured tasks
• Assigns owners automatically
• Tracks deadlines
• Sends reminders
• Escalates if delayed

---

Tech Stack

• LLM: GPT-4o
• Backend: Node.js
• Slack API: Bolt.js
• RAG: ChromaDB
• Memory: mem0
• Tool Integration: MCP (GitHub + Notion)

---

Installation

git clone https://github.com/your-repo
cd project
npm install
npm run dev

---

Project Structure

src/
├── agents/          # Multi-agent logic
├── orchestrator/    # Workflow controller
├── rag/             # Context retrieval
├── memory/          # Long-term memory
├── mcp/             # Tool integrations
├── slack/           # Slack interface

---

Demo

3-minute walkthrough should include:

• Input to workflow execution
• Agent coordination
• Automatic error recovery
• Audit logging

---

Submission Checklist

• GitHub Repository
• README
• Pitch Video
• Architecture Diagram
• Impact Model

---

Future Improvements

• Add enterprise workflows (HR, Finance, Operations)
• Predictive failure detection
• Improved agent collaboration logic
• Workflow visualization dashboard

---

Conclusion

This project transforms a chatbot into an autonomous enterprise worker:

• From reactive to proactive systems
• From single-agent to multi-agent architecture
• From manual workflows to autonomous execution