Multi-Agent Ad Investment Framework

LangGraph-based multi-agent system for Amazon & Flipkart ad budget optimization.

Architecture

Campaign Data
     │
     ▼
┌─────────────────┐
│ Market Analyst  │  → ACoS, ROAS, CTR analysis per channel
└────────┬────────┘
         ▼
┌─────────────────┐
│ Budget Optimizer│  → ROAS-weighted reallocation across 6 channels
└────────┬────────┘
         ▼
┌─────────────────┐
│ Bid Strategy    │  → CPC/CPM bids, dayparting, match type shifts
└────────┬────────┘
         ▼
┌─────────────────┐
│ Risk Guard      │  → ACoS cap, overspend, inventory stockout check
└────────┬────────┘
         ▼
┌─────────────────┐
│ Execution Plan  │  → Final investment brief + monitoring triggers
└─────────────────┘

Setup

pip install langgraph langchain-groq langchain-core rich tabulate

Get a free Groq API key at: https://console.groq.com

export GROQ_API_KEY=your_key_here

Run

# Interactive demo (recommended for presentations)
python demo.py

# Direct run with default data
python agents.py

Demo Scenarios

#	Scenario	Budget	Season
1	Pre-Diwali Push	₹5,00,000	High Competition
2	Post-Sale Cleanup	₹2,00,000	Low Efficiency
3	New Product Launch	₹1,50,000	Zero History
4	Custom Input	User-defined	Any

Channels Covered

Amazon: Sponsored Products · Sponsored Brands · DSP Display
Flipkart: Product Ads · PLA Ads · Brand Ads

What Each Agent Does

Agent	Input	Output
Market Analyst	Raw KPIs	ROAS ranking, red flags
Budget Optimizer	Analysis + total budget	New spend per channel
Bid Strategy	Budget plan + competition	CPC/CPM per channel, dayparting
Risk Guard	Full plan	Risk table + GO/HOLD/REVISE
Execution Planner	All outputs	Final brief + 3 actions this week

Extending

To plug in real data, replace CAMPAIGN_DATA in agents.py with your Amazon Ads API / Flipkart Ads API response.

To add a new agent, add a node to the StateGraph in build_graph().