Skip to content

Vritika22Mandapaka/Multi_Agent_Business_System

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AI Multi-Agent Business Decision System

Project Overview

This project is an AI-powered Multi-Agent Business Decision System built for FE524 – Prompt Engineering Lab for Business Applications at Stevens Institute of Technology.

The system helps evaluate a business idea by simulating how real companies make strategic decisions.

Instead of relying on separate teams for:

  • Market Research
  • Technical Feasibility
  • Financial Analysis

this system uses multiple AI agents to perform all three analyses automatically and generates one final business recommendation.

The final output includes:

  • Go / No-Go decision
  • Confidence Score
  • Market analysis
  • Technical feasibility report
  • Financial viability report
  • Consolidated risk list
  • Final recommendation

The project now also includes a Streamlit web interface for live demo presentation and user-friendly interaction.


Business Problem

When companies plan major strategic decisions such as:

  • launching a new product
  • entering a new market
  • building a startup

they usually need multiple teams working separately.

This creates problems like:

  • slow decision-making
  • inconsistent assumptions
  • delayed execution
  • fragmented recommendations

This project solves that by using AI agents working together inside a single LangGraph workflow.


System Architecture

The system is built using:

  • LangGraph → multi-agent workflow orchestration
  • LangChain → prompt templates and model wrappers
  • OpenAI GPT-5 nano → primary LLM
  • llama.cpp (fallback-ready) → local LLM backup
  • PyMuPDF → PDF input parsing
  • Python tools → calculator + web search simulation
  • Streamlit → web-based interactive UI for demo and presentation

Agent Workflow

Agent 1 — Research Agent

Responsible for:

  • Market trends
  • Competitor analysis
  • Demand estimation
  • External risks

Tool used:

  • Web Search Tool

Output:

  • Market opportunity summary
  • Competitor analysis
  • Demand assessment
  • External risk flags

Agent 2 — Technology Stack Agent

Responsible for:

  • Technical feasibility
  • Stack recommendation
  • Development effort estimate
  • Technical risks

Output:

  • Feasibility verdict
  • Recommended tech stack
  • Development effort
  • Technical risks

Agent 3 — Finance Agent

Responsible for:

  • Startup cost estimation
  • ROI analysis
  • Break-even calculation
  • Payback period
  • Financial risks

Tool used:

  • Calculator Tool

Output:

  • Cost breakdown
  • ROI projection
  • Break-even analysis
  • Financial risk flags

Retry Agent

Responsible for:

  • Re-analysis when confidence is low

Used when:

  • confidence score is too low
  • high-risk signals are detected
  • business feasibility is uncertain

This improves reliability and matches real multi-agent decision systems.


Synthesis Layer

Responsible for:

  • Combining all agent outputs

Final output:

  • Final Verdict
  • Overall Confidence Score
  • Research Summary
  • Technology Summary
  • Finance Summary
  • Consolidated Risk List
  • Final Recommendation

Features Implemented

Core Features

  • PDF input support
  • TXT input support
  • GPT-5 nano integration
  • OpenAI API connection
  • Local LLM fallback-ready structure
  • LangGraph multi-agent execution

Tool Usage

  • Web Search Tool for Research Agent
  • Calculator Tool for Finance Agent

Advanced Features

  • Parallel/branching workflow
  • Conditional retry routing
  • Low-confidence retry node
  • Shared typed state
  • Final synthesis node

Evaluation Features

  • Rubric-based scoring
  • Consistency check across multiple runs
  • Confidence validation
  • Structured output evaluation

UI Features

  • Streamlit dashboard
  • PDF/TXT upload support
  • Manual business idea input
  • Real-time final report generation
  • Agent-wise output display
  • Evaluation metrics display
  • Consistency score visualization

Project Structure

Multi_Agent_Busisys/
│
├── app/
│   ├── main.py
│   ├── graph.py
│   ├── state.py
│   ├── llm_client.py
│   └── __init__.py
│
├── agents/
│   ├── research_agent.py
│   ├── tech_agent.py
│   ├── finance_agent.py
│   ├── retry_agent.py
│   ├── synthesis_agent.py
│   └── __init__.py
│
├── tools/
│   ├── calculator.py
│   ├── web_search.py
│   └── __init__.py
│
├── evals/
│   ├── rubric_eval.py
│   ├── consistency_check.py
│   └── __init__.py
│
├── input/
│   └── sample_business_idea.txt
│
├── output/
│
├── streamlit_app.py
│
├── requirements.txt
├── .env
├── .gitignore
└── README.md

Installation

Step 1 — Create Virtual Environment

py -m venv .venv
.\.venv\Scripts\activate

Step 2 — Install Requirements

pip install -r requirements.txt

Step 3 — Add Environment Variables

Create .env

OPENAI_API_KEY=your_api_key_here
LLM_BACKEND=openai
OPENAI_MODEL=gpt-5-nano

Running the Project

Terminal Version

py -m app.main

Then enter:

input/sample_business_idea.txt

or any custom PDF/TXT business idea file.


Streamlit Web App Version

streamlit run streamlit_app.py

This opens a browser-based dashboard where users can:

  • Upload PDF/TXT files
  • Type business ideas manually
  • Run complete multi-agent analysis
  • View final reports
  • See all agent outputs
  • Check rubric scores
  • View consistency check results

This version is ideal for final project presentation and professor demo.


Evaluation Metrics

Rubric-Based Scoring

Each agent is scored based on required outputs.

Example:

  • Research Agent → market, competitors, demand, risk
  • Finance Agent → cost, ROI, break-even, payback

Consistency Check

The same input is run multiple times.

The system compares:

  • recommendation quality
  • confidence stability
  • repeated conclusions

This validates output consistency.

Example result:

Consistency Score: 87.5%

Example Final Output

Final Verdict:
Proceed with campus-first MVP

Confidence Score:
7.5 / 10

Break-even:
55,437 orders/month

ROI:
523%

Payback Period:
18.1 months

Course Concepts Used

Course Topic Implementation
Few-shot prompting Agent prompt design
Chain-of-thought Synthesis reasoning
Tool use Calculator + Web Search
Open models llama.cpp fallback
LangGraph Multi-agent orchestration
Evaluation Rubric + Consistency
Prompt Engineering Zero-shot + Few-shot + CoT
Streamlit UI Live interactive demo

Authors

  • Sri Harsha Chinta
  • Kamakshi Padma Vritika Manadapaka
  • Siddharth Raju Vysyaraju
  • Sri Lasya Siripurapu

Final Note

This project demonstrates how modern AI agents can support real-world business strategy decisions using:

  • structured reasoning
  • tool calling
  • evaluation systems
  • multi-agent orchestration
  • interactive Streamlit presentation layer

rather than simple single-prompt LLM outputs.

It is designed to reflect realistic enterprise decision-making workflows and present them in a professor-level production-ready format.

About

AI-powered Multi-Agent Business Decision System using LangGraph, GPT-5 nano, and Streamlit to analyze business ideas through Research, Technology, Finance, and Synthesis agents. The system provides market analysis, technical feasibility, financial evaluation, risk assessment, confidence scoring, and final go/no-go recommendations with checks etc

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages