Skip to content

19Vermouth/MergerMind_AI

BranchesTags

Repository files navigation

DealSense AI — AI-Powered M&A Intelligence Platform

Build status: CI License

What is DealSense AI?

DealSense AI is a production-grade, end-to-end platform that automates Mergers & Acquisitions due diligence by combining data engineering, ML, NLP, financial simulation, and LLM intelligence into a unified pipeline.

It takes a proposed deal (e.g., "Adobe acquiring Figma") and returns:

  • ✅ Deal success probability (ML model)
  • 📰 Market sentiment score (FinBERT + NewsAPI)
  • 🎲 50,000 simulated outcomes (Monte Carlo NPV/IRR/VaR)
  • 🤖 AI recommendation — PROCEED | NEGOTIATE | REJECT
  • 📋 Executive summary (LLM-generated)

Architecture

Architecture Diagram

flowchart TB
    subgraph Sources["Data Sources"]
        SCRAPE[Scrapy Spiders]
        NEWS[NewsAPI]
    end

    subgraph Bronze["Bronze Layer - MinIO (Raw)"]
        RAW_HTML[Raw HTML / JSON]
    end

    subgraph Airflow["Apache Airflow (Orchestration)"]
        DAG["DAG: ingest → transform → quality"]
    end

    subgraph Silver["Silver Layer - PostgreSQL Staging"]
        STG_DEALS[stg_ma_deals]
        STG_NEWS[stg_news_articles]
    end

    subgraph Gold["Gold Layer - dbt Transformations"]
        DIM["dim_companies<br/>dim_industries"]
        FACT["fact_ma_deals<br/>fact_news_sentiment"]
    end

    subgraph ML["ML Pipeline"]
        FEATURES[Feature Engineering]
        TRAIN["ML Training<br/>XGBoost / RF"]
        MLFLOW[MLflow Tracking]
    end

    subgraph Analysis["Analysis Engine"]
        SENTIMENT[FinBERT Sentiment]
        MONTE_CARLO["Monte Carlo<br/>50,000 sims"]
        SCORER[Deal Scorer]
        LLM["Groq / OpenRouter<br/>Recommendation"]
    end

    subgraph API["FastAPI"]
        ENDPOINTS["POST /analyze-deal<br/>GET /deal/:id<br/>GET /health"]
    end

    subgraph UI["Streamlit Dashboard"]
        PAGES["Overview | Explorer | Intelligence<br/>Risk | AI Report | Model Performance"]
    end

    Sources --> Bronze
    Bronze --> Airflow
    Airflow --> Silver
    Silver --> Gold
    Gold --> FEATURES
    FEATURES --> TRAIN
    TRAIN --> MLFLOW

    NEWS --> SENTIMENT
    SENTIMENT --> SCORER
    MONTE_CARLO --> SCORER
    SCORER --> LLM
    LLM --> API
    API --> UI

    SCRAPE --> RAW_HTML
    NEWS --> RAW_HTML
Loading

Data Flow Summary

Layer Technology Purpose
Bronze MinIO (S3) Raw HTML, JSON, API responses
Orchestration Airflow DAG scheduling, retries, alerting
Silver PostgreSQL Staging Parsed, validated raw data
Gold dbt Star-schema warehouse transformations
ML scikit-learn / XGBoost Deal success prediction
NLP FinBERT (HuggingFace) News sentiment scoring
Simulation NumPy / SciPy 50,000 Monte Carlo iterations
LLM Groq / OpenRouter / Gemini Recommendation + executive summary
API FastAPI + Pydantic REST endpoints
Dashboard Streamlit + Plotly Interactive visualizations
Tracking MLflow Experiment versioning
Quality Great Expectations Schema + null rate validation

Quick Start

Requires Docker and Docker Compose v2+

# 1. Clone the repository
git clone https://github.com/dealsense/dealsense-ai.git
cd dealsense-ai

# 2. Configure environment
cp .env.example .env
# Edit .env and add your API keys:
#   - NEWS_API_KEY (https://newsapi.org)
#   - GROQ_API_KEY  (https://console.groq.com)
#   - GEMINI_API_KEY (https://aistudio.google.com)

# 3. Launch entire stack
make up

# 4. Initialize database schema
make init-db

# 5. Seed sample data
make seed-db

# 6. Open the dashboard
open http://localhost:8501

# 7. Analyze a deal via API
curl -X POST http://localhost:8000/analyze-deal \
  -H "Content-Type: application/json" \
  -d '{"acquirer": "Microsoft", "target": "GitHub", "industry": "Software", "deal_value_usd": 7500000000}'

Five commands to full stack. That's it.

Project Structure

dealsense-ai/
├── .env.example                 # All environment variables
├── Makefile                     # Automation commands (make up, make test, etc.)
├── README.md                    # This file
│
├── docker-compose.yml          # Full stack: Postgres, MinIO, Airflow, MLflow,
│                               #   FastAPI, Streamlit
│
├── docker/                     # Docker-related files
│   └── init-scripts/
│       ├── init.sql            # Schema: raw, staging, mart, ml, metadata
│       └── seed.sql            # 50 sample M&A deals + news articles
│
├── docker/Dockerfile.api       # FastAPI container
├── docker/Dockerfile.dashboard # Streamlit container
├── docker/Dockerfile.scraper   # Scrapy container
└── docker/Dockerfile.dbt       # dbt container
│
├── requirements/               # Python dependencies
│   ├── base.txt
│   ├── dev.txt
│   └── prod.txt
│
├── src/                        # Python source code
│   ├── ingestion/              # Bronze → MinIO loader
│   │   ├── __init__.py
│   │   ├── loader.py          # MinIO and Postgres ingestion
│   │   └── schemas.py         # Pydantic models for deal data
│   │
│   ├── features/              # Feature engineering
│   │   ├── __init__.py
│   │   └── feature_engineering.py  # Industry similarity, multiples, premiums
│   │
│   ├── models/                # ML training and prediction
│   │   ├── __init__.py
│   │   ├── train.py           # Training pipeline with MLflow
│   │   └── predict.py         # Prediction with confidence intervals
│   │
│   ├── simulation/            # Monte Carlo engine
│   │   ├── __init__.py
│   │   └── monte_carlo.py     # 50,000 simulations, NPV/IRR/VaR
│   │
│   ├── scoring/               # Deal scoring engine
│   │   ├── __init__.py
│   │   └── deal_scorer.py     # Combines ML + sentiment + simulation
│   │
│   ├── llm/                   # LLM integration
│   │   ├── __init__.py
│   │   ├── providers.py       # Groq / OpenRouter / Gemini wrappers
│   │   └── recommendation_engine.py  # Prompt builder + response parser
│   │
│   ├── api/                   # FastAPI application
│   │   ├── __init__.py
│   │   ├── main.py            # App entry, CORS, middleware
│   │   ├── config.py          # Settings from environment
│   │   ├── routes/
│   │   │   ├── __init__.py
│   │   │   ├── deals.py       # POST /analyze-deal, GET /deal/{id}
│   │   │   └── health.py      # GET /health
│   │   └── models.py          # Pydantic request/response schemas
│   │
│   └── dashboard/             # Streamlit dashboard
│       ├── app.py             # Main app + sidebar navigation
│       └── pages/
│           ├── 1_overview.py
│           ├── 2_deal_explorer.py
│           ├── 3_news_intelligence.py
│           ├── 4_risk_analysis.py
│           └── 5_ai_report.py
│
├── dags/                      # Apache Airflow DAGs
│   ├── ma_ingestion_dag.py    # Main ingestion pipeline
│   ├── ml_training_dag.py     # Weekly model retraining
│   └── news_sentiment_dag.py  # Daily news fetching
│
├── dbt/                       # dbt Core project
│   ├── dbt_project.yml
│   ├── profiles.yml
│   ├── packages.yml
│   └── models/
│       ├── raw/
│       │   └── sources.yml    # Source definitions
│       ├── staging/
│       │   ├── stg_ma_deals.sql
│       │   └── stg_news_sentiment.sql
│       ├── mart/
│       │   ├── dim_companies.sql
│       │   ├── dim_industries.sql
│       │   ├── fact_ma_deals.sql
│       │   ├── fact_news_sentiment.sql
│       │   └── deal_analysis_results.sql
│       └── ml/
│           └── feature_deal_model.sql
│
├── scraping/                  # Scrapy project
│   ├── scrapy.cfg
│   └── spiders/
│       ├── __init__.py
│       └── ma_deals_spider.py  # M&A deal scraper
│
├── tests/                     # pytest + coverage
│   ├── conftest.py            # Shared fixtures
│   ├── test_monte_carlo.py    # 50k simulation tests
│   ├── test_scorer.py         # Recommendation logic tests
│   ├── test_api.py            # API endpoint tests
│   ├── test_features.py       # Feature engineering tests
│   └── test_sentiment.py      # NLP pipeline tests
│
├── notebooks/                 # Jupyter exploration
│   ├── EDA_ma_deals.ipynb
│   ├── model_exploration.ipynb
│   └── monte_carlo_analysis.ipynb
│
├── .github/
│   └── workflows/
│       └── ci.yml             # GitHub Actions pipeline
│
└── .pre-commit-config.yaml   # pre-commit hooks

API Contract

POST /analyze-deal

Request:

{
  "acquirer": "Microsoft",
  "target": "GitHub",
  "industry": "Software",
  "deal_value_usd": 7500000000
}

Response:

{
  "deal_id": "550e8400-e29b-41d4-a716-446655440000",
  "acquirer": "Microsoft",
  "target": "GitHub",
  "deal_value_usd": 7500000000,
  "success_probability": 0.82,
  "sentiment_score": 0.64,
  "expected_npv": 2400000000,
  "probability_positive_npv": 0.78,
  "var_95": -850000000,
  "irr_median": 0.18,
  "recommendation": "PROCEED",
  "confidence": "HIGH",
  "executive_summary": "Based on analysis of 847 comparable deals...",
  "risk_factors": ["Integration complexity", "Regulatory approval uncertainty"],
  "key_metrics": {
    "ev_revenue_multiple": 25.0,
    "premium_paid": 0.49,
    "industry_success_rate": 0.71
  },
  "simulation_percentiles": {
    "p10": 800000000,
    "p25": 1500000000,
    "p50": 2400000000,
    "p75": 3500000000,
    "p90": 4800000000
  }
}

GET /health

{"status": "healthy", "version": "1.0.0", "services": {"postgres": "up", "minio": "up"}}

Dashboard Screens

Page Description
Executive Overview KPI cards, deal probability gauge, recommendation card
Deal Explorer Filterable table of historical M&A deals with multiples
News Intelligence Sentiment timeline, article headlines, sector trends
Monte Carlo Risk NPV histogram, IRR distribution, VaR chart, percentile table
AI Report Full LLM-generated executive summary + risk commentary
Model Performance ROC curve, feature importance, training metrics

Monte Carlo Model

Simulates 50,000 deal outcomes using:

Variable Distribution Range
Revenue synergies Log-normal -5% to +40% of deal value
Cost synergies Normal 0 to +25% of deal value
Integration costs Gamma $50M to $500M
Market volatility Historical VaR 1-year rolling
Discount rate Uniform 8% to 14%
Regulatory delay Poisson 0 to 18 months

Output metrics: Expected NPV, IRR distribution, P(NPV > 0), VaR (95%), percentile analysis

ML Pipeline

Models: Logistic Regression, Random Forest, XGBoost

Features:

  • Industry similarity score
  • Deal size (log-transformed)
  • Premium paid (vs. 30-day moving avg)
  • EV/Revenue multiple
  • EV/EBITDA multiple
  • Regulatory complexity score
  • Market volatility at announcement
  • Historical sector success rate
  • News sentiment score (0-1)

Target: Binary — successful integration (1) vs. failed (0)

Tracking: All experiments logged to MLflow with:

  • ROC-AUC, Precision, Recall, F1
  • Feature importance
  • Confusion matrix
  • Training curves

Data Quality (Great Expectations)

Automated validation checks on every pipeline run:

  • Schema enforcement (expected columns + types)
  • Null rate thresholds (< 5% for critical fields)
  • Value ranges (deal_value_usd > 0, sentiment_score ∈ [0,1])
  • Row count sanity checks
  • Referential integrity (foreign keys)

Development

# Install dependencies locally (not in Docker)
pip install -r requirements/dev.txt

# Run tests
make test

# Lint and format
make lint
make format

# Run a specific DAG manually
docker compose exec airflow-webserver airflow dags trigger ma_ingestion

# Open MLflow
open http://localhost:5001

# Tail Airflow logs
make logs-airflow

# Reset everything
make clean && make up && make init-db && make seed-db

Contributing

  1. Fork → Branch → Commit → PR
  2. Run make lint && make test before opening PR
  3. Maintain 80%+ test coverage
  4. Use type hints everywhere
  5. Update this README if you add features

Environment Variables

See .env.example for all configuration options.

Variable Description Required
POSTGRES_* Database connection Yes
MINIO_* S3 storage Yes
NEWS_API_KEY News fetching Yes
GROQ_API_KEY Primary LLM Yes
OPENROUTER_API_KEY Fallback LLM No
GEMINI_API_KEY Fallback LLM No
MODEL_WEIGHT_* Scoring weights No
MONTE_CARLO_SIMULATIONS Number of sims No

License

MIT License — see LICENSE for details.

About

DealSense AI is an end-to-end M&A intelligence platform that combines machine learning, sentiment analysis, Monte Carlo simulation, and LLMs to automate due diligence and generate data-driven deal recommendations and executive summaries.

Resources

Readme
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages