███╗   ███╗███████╗███████╗████████╗███╗   ███╗██╗███╗   ██╗██████╗
████╗ ████║██╔════╝██╔════╝╚══██╔══╝████╗ ████║██║████╗  ██║██╔══██╗
██╔████╔██║█████╗  █████╗     ██║   ██╔████╔██║██║██╔██╗ ██║██║  ██║
██║╚██╔╝██║██╔══╝  ██╔══╝     ██║   ██║╚██╔╝██║██║██║╚██╗██║██║  ██║
██║ ╚═╝ ██║███████╗███████╗   ██║   ██║ ╚═╝ ██║██║██║ ╚████║██████╔╝
╚═╝     ╚═╝╚══════╝╚══════╝   ╚═╝   ╚═╝     ╚═╝╚═╝╚═╝  ╚═══╝╚═════╝

Autonomous Meeting Intelligence Agent System

Turn any meeting transcript into fully executed workflows — zero manual follow-up

---

🎯 What is MeetMind?

MeetMind is a multi-agent AI system that takes a raw meeting transcript and autonomously executes the entire post-meeting workflow — no human follow-up required.

Most tools (Otter.ai, Fireflies, Notion AI) stop at summarization — they produce a document someone still has to act on. MeetMind acts.

Raw Transcript  ──▶  Extract Tasks  ──▶  Resolve Owners  ──▶  Create JIRA Tasks
                                                                      │
Audit Log  ◀──  Escalate Stalls  ◀──  Monitor Progress  ◀──  Send Summary

In under 90 seconds, MeetMind:

• Extracts every action item with context and deadlines
• Resolves who owns each task (with reasoning, not guessing)
• Creates tasks in your project tracker automatically
• Sends a structured summary to all participants
• Monitors completion and escalates stalls to managers
• Logs every decision with a full, immutable audit trail

---

🏆 Hackathon Context

Field	Detail
Event	ET AI Hackathon 2026 — Avataar.ai × Economic Times
Track	Problem Statement 2: Agentic AI for Autonomous Enterprise Workflows
Submission Phase	Phase 2 — Build Sprint
Deadline	27 March 2026

Why PS 2?

PS 2 is the flagship track of this hackathon. It demands the highest bar:

• ≥5 autonomous sequential steps ✅ (MeetMind: 8 steps)
• Live error-recovery scenario ✅ (2 branch points demonstrated)
• Full audit trail of every agent decision ✅
• Minimal human involvement ✅ (humans only resolve explicit ambiguities)

---

🤖 Agent Architecture

MeetMind uses 6 specialized agents orchestrated by a LangGraph state machine.

┌─────────────────────────────────────────────────────────────────────┐
│                        MEETMIND AGENT GRAPH                         │
│                                                                     │
│   ┌─────────────┐                                                   │
│   │ TRANSCRIPT  │  (text / audio / file upload)                     │
│   └──────┬──────┘                                                   │
│          ▼                                                          │
│   ┌─────────────────┐                                               │
│   │  ORCHESTRATOR   │  LangGraph State Machine                      │
│   └────────┬────────┘                                               │
│            │                                                        │
│     ┌──────▼──────┐                                                 │
│     │  EXTRACTOR  │  Parse action items, decisions, deadlines       │
│     │   AGENT     │  Model: Llama 3.3 70B (Groq)                   │
│     └──────┬──────┘                                                 │
│            │                                                        │
│     ┌──────▼──────────┐    ┌─────────────────┐                     │
│     │  OWNER RESOLVER │───▶│ AMBIGUOUS FLAG  │──▶ Human UI         │
│     │     AGENT       │    │ (never guesses) │                     │
│     └──────┬──────────┘    └─────────────────┘                     │
│            │                                                        │
│     ┌──────▼──────┐    ┌──────────────┐                            │
│     │    TASK     │───▶│ JIRA FAILURE │──▶ Retry × 3 ──▶ Notion   │
│     │   CREATOR   │    │   RECOVERY   │    Fallback                │
│     └──────┬──────┘    └──────────────┘                            │
│            │                                                        │
│     ┌──────▼──────┐                                                 │
│     │    COMMS    │  Send structured summary to all participants    │
│     │    AGENT    │                                                 │
│     └──────┬──────┘                                                 │
│            │                                                        │
│     ┌──────▼──────┐    ┌─────────────────┐                         │
│     │   MONITOR   │───▶│ STALL DETECTED  │──▶ Escalation to        │
│     │  ESCALATOR  │    │ (T+24/48/72h)   │    manager with context │
│     └──────┬──────┘    └─────────────────┘                         │
│            │                                                        │
│     ┌──────▼──────┐                                                 │
│     │ AUDIT TRAIL │  SQLite — every action, timestamp, reasoning   │
│     └─────────────┘                                                 │
└─────────────────────────────────────────────────────────────────────┘

Agent Responsibilities

Agent	Responsibility	LLM Model	Error Handling
🎯 Orchestrator	State machine; routes between agents	—	Catches all exceptions
📋 Extractor	Parses transcript → structured JSON	Llama 3.3 70B	Retry; Gemini fallback
👤 Owner Resolver	Maps owners; flags ambiguity	Llama 3.3 70B	Never guesses
🔧 Task Creator	Creates JIRA tasks; retry + Notion fallback	Llama 3.1 8B	3× retry → Notion
📧 Comms Agent	Drafts + sends meeting summary	Llama 3.1 8B	Partial delivery handling
🚨 Monitor & Escalator	Polls tasks; escalates stalls	Llama 3.3 70B	Escalation chain

---

⚡ 8-Step Autonomous Workflow

Step 1  │ Orchestrator   │ Initialize state, validate transcript, start audit log
Step 2  │ Extractor      │ LLM extracts action items with context + deadlines
Step 3* │ Owner Resolver │ Map owners → team; FLAG ambiguities (never guess)
Step 4  │ Task Creator   │ Create JIRA tasks; retry on failure; Notion fallback
Step 5  │ Comms Agent    │ Send structured summary to all participants
Step 6* │ Monitor Agent  │ Poll at T+24/48/72h; escalate stalls to manager
Step 7  │ Orchestrator   │ Compile workflow completion report
Step 8  │ Orchestrator   │ Persist immutable audit log; close state

* Steps 3 and 6 contain live branching logic demonstrated during judging.

---

🛠️ Tech Stack

Agent Framework:    LangGraph 0.2+
Primary LLM:        Groq / Llama 3.3 70B
Small Model:        Groq / Llama 3.1 8B
Fallback LLM:       Google Gemini 2.0 Flash
Backend:            FastAPI
Frontend/Demo:      Streamlit
Audit Storage:      SQLite
Task Tracker:       JIRA REST API (mock)
Scheduler:          APScheduler
Language:           Python 3.11+

LLM Cost-Efficiency Routing

Complex tasks  ──▶  Llama 3.3 70B  (Groq)
Simple tasks   ──▶  Llama 3.1 8B   (Groq, 10× faster)
Rate limit hit ──▶  Gemini 2.0 Flash (auto-switch)

---

📁 Project Structure

meetmind/
├── agents/
│   ├── orchestrator.py
│   ├── extractor.py
│   ├── owner_resolver.py
│   ├── task_creator.py
│   ├── comms_agent.py
│   └── monitor_escalator.py
├── tools/
│   ├── jira_client.py
│   ├── notion_client.py
│   ├── smtp_mock.py
│   └── audit_logger.py
├── core/
│   ├── state.py
│   ├── graph.py
│   ├── llm_router.py
│   └── config.py
├── api/
│   └── main.py
├── ui/
│   └── app.py
├── data/
│   ├── sample_transcripts/
│   │   ├── scenario_1_meeting_to_action.txt
│   │   ├── scenario_2_sla_breach.txt
│   │   └── scenario_3_ambiguous_owner.txt
│   └── team_roster.json
├── docs/
│   └── MeetMind_Architecture.docx
├── requirements.txt
├── .env.example
└── README.md

---

🚀 Setup & Run

Prerequisites

• Python 3.11+
• Groq API Key — free
• Google AI Studio Key — free

1. Clone & Install

git clone https://github.com/Nimeshdev1/meetmind.git
cd meetmind
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt

2. Configure Environment

cp .env.example .env
# Add your GROQ_API_KEY and GOOGLE_API_KEY to .env

3. Run

streamlit run ui/app.py

Opens at http://localhost:8501

---

🎬 Demo Scenarios

Scenario	What it demonstrates
scenario_1_meeting_to_action	Full pipeline + ambiguity flagging + escalation
scenario_2_sla_breach	SLA breach detection + task rerouting
scenario_3_ambiguous_owner	Multiple ambiguities + human resolution UI

---

📊 Business Impact

Metric	Before	After	Change
Post-meeting admin time	25 min	2 min	−92%
Task ownership clarity	40%	98%	+145%
Task creation lag	Hours	<90 sec	−99%
Audit trail coverage	0%	100%	Full

Annual saving: ₹6.13 Crore/year for a 500-person org

---

📋 Evaluation Rubric Mapping

Dimension	Weight	Our Implementation
Autonomy Depth	30%	8 steps, 0 human touchpoints, 2 error-recovery branches
Multi-Agent Design	20%	6 agents, LangGraph state machine, single-responsibility
Technical Creativity	20%	LLM routing 70B/8B/Gemini, checkpointing, ambiguity-first
Enterprise Readiness	20%	SQLite audit trail, exponential backoff, no silent failures
Impact Quantification	10%	₹6.13Cr/year with stated assumptions

---

🙏 Acknowledgements

Built for ET AI Hackathon 2026 by Avataar.ai × Economic Times.

• LangGraph
• Groq
• Google AI Studio
• FastAPI
• Streamlit

---

MeetMind — ET AI Hackathon 2026 · PS2: Agentic AI for Autonomous Enterprise Workflows