SentinelAI - Real-Time Due Diligence Intelligence

AI-powered due diligence system with 8 specialized agents. Real-time risk analysis across legal, financial, governance, reputation & operations.

Comment

Inspiration

Due diligence is critical for investors, businesses, and decision-makers, but it's incredibly time-consuming. Analysts spend hours manually searching through news articles, financial reports, legal filings, and social media to assess risks. We wanted to democratize this process by building an AI system that could perform comprehensive due diligence in minutes instead of hours, making professional-grade risk analysis accessible to everyone.

What it does

SentinelAI is an AI-powered multi-agent system that performs automated due diligence analysis on companies, startups, and entities. It:

  • Executes real-time web searches across 15 targeted queries using You.com's Search API
  • Analyzes risk across 5 categories: Legal (30%), Financial (25%), Reputation (20%), Governance (15%), and Operational (10%)
  • Detects contradictions in source data to identify conflicting information
  • Generates comprehensive reports with risk scores (0-10), confidence levels, and actionable recommendations
  • Provides source citations for every finding, ensuring transparency and verifiability
  • Displays results in a beautiful, minimalist dark-themed interface

The system processes 75+ web sources per analysis and delivers results in under 60 seconds.

How we built it

Architecture: We implemented a multi-agent pipeline with 8 specialized AI agents:

  1. Orchestrator - Coordinates the entire analysis workflow
  2. Query Expander - Generates 15 targeted search queries across risk categories
  3. Search Agent - Executes concurrent web searches via You.com API
  4. Evidence Extractor - Parses search results and extracts structured risk signals
  5. Risk Categorizer - Classifies evidence into risk types (legal, financial, etc.)
  6. Risk Scorer - Calculates weighted risk scores using the formula:

$$\text{Risk Score} = \sum_{i=1}^{5} w_i \cdot s_i$$

where $w_i$ are category weights and $s_i$ are severity scores

  1. Contradiction Detector - Identifies conflicting claims across sources
  2. Report Generator - Produces final analysis with recommendations

Tech Stack:

  • Backend: FastAPI with async/await for concurrent processing
  • Search: You.com Search API for real-time web intelligence
  • Database: SQLite with SQLAlchemy for analysis history
  • Caching: Redis (optional) for performance optimization
  • Frontend: Vanilla JavaScript with modern CSS (no frameworks)

Key Implementation Details:

  • Async processing enables 15 concurrent searches, reducing analysis time by 10x
  • Weighted scoring model based on industry-standard risk assessment frameworks
  • Fallback extraction when advanced NLP isn't available
  • Graceful degradation - system works even without Redis caching

Accomplishments that we're proud of

✨ Built a production-ready system in record time - Full multi-agent architecture with beautiful UI

🎯 Accurate risk assessment - Tested on real companies (Stripe, OpenAI) and results aligned with known public information

⚡ Performance optimization - Concurrent processing delivers results in <60 seconds despite analyzing 75+ sources

🎨 Beautiful UX - Created a minimalist, dark-themed interface that's both functional and elegant

🔍 Transparency - Every finding is linked to original sources with citations

📊 Comprehensive analysis - 5 risk categories, contradiction detection, confidence scoring, and actionable recommendations

What we learned

Technical Learnings:

  • Multi-agent architectures are powerful for complex workflows - each agent has a single responsibility
  • Async/await in Python dramatically improves I/O-bound operations (15x speedup)
  • API documentation can be misleading - always test thoroughly and check actual responses
  • Graceful degradation is crucial - system should work even when optional services fail
  • Type hints and Pydantic schemas catch bugs early and improve code quality

Domain Learnings:

  • Due diligence requires balancing multiple risk factors with different weights
  • Contradiction detection is critical - conflicting information often signals deeper issues
  • Source credibility matters - not all web sources are equally reliable
  • Confidence scoring helps users understand analysis limitations

Product Learnings:

  • Simple, clean UI beats feature-rich complexity
  • Real-time feedback (loading states, progress) improves perceived performance
  • Export functionality is essential for professional tools
  • Demo mode enables testing without API keys

What's next for SentinelAI - Real-Time Due Diligence Intelligence

Short-term (Next Month):

  • 🧪 Add comprehensive unit and integration tests
  • 📈 Implement historical risk tracking and trend analysis
  • 🔔 Add email/Slack notifications for watchlist alerts
  • 📄 Generate PDF reports for professional sharing

Medium-term (3-6 Months):

  • 🤖 Integrate advanced NLP models for better evidence extraction
  • 🌍 Multi-language support for international entities
  • 📊 Custom risk scoring models (let users define their own weights)
  • 🔗 API integrations with CRM systems (Salesforce, HubSpot)

Long-term Vision:

  • 🎯 Sector-specific analysis (fintech, healthcare, crypto)
  • 🤝 Collaborative features for team-based due diligence
  • 📱 Mobile app for on-the-go analysis
  • 🧠 Machine learning to improve scoring accuracy over time
  • 🌐 Enterprise version with SSO, audit logs, and compliance features

Moonshot Goal: Become the go-to platform for automated due diligence, helping investors, businesses, and analysts make better decisions faster.

Built With

Share this project:

Updates