D3 · Deep Due Diligence

D3 is an AI agent powered by Apify that mines forums and legal dockets to create a Risk Dossier, uncovering corporate operational cracks missed by surface-level due diligence.

Comment

D3 · Deep Due Diligence

Auditing corporate integrity by mining the "digital basement."

🚢 Inspiration

We all remember the OceanGate Titan tragedy of June 18, 2023. In the aftermath, the world wondered: How did they not see the signs? The truth is, the signs weren't just visible; they were scattered across the internet.

Years before the implosion, whistleblower reports existed in legal dockets, expert warnings were posted on niche engineering forums, and Glassdoor reviews from former employees described a "toxic safety culture" and "expired materials." But for the average passenger or investor, these "scent trails" were buried under layers of polished marketing and SEO.

D3 (Deep Due Diligence) was built to ensure that "innovation" never again serves as a mask for negligence. We provide a dedicated AI "private investigator" for anyone who needs to verify if a company’s reality matches its claims.

🔍 What it does

D3 is an AI-powered agent that identifies, scrapes, and analyzes the "unstructured truth" about an organization. While traditional due diligence focuses on financial spreadsheets, D3 dives into:

  • Technical Forums (Reddit/Specialized Boards): Finding expert discussions on product flaws.

  • Employee Sentiment (Glassdoor): Detecting internal red flags like safety-cutting or high turnover in critical roles.

  • Legal & Regulatory Records: Identifying forgotten lawsuits or safety violations.

The app performs a Divergence Analysis, specifically highlighting the gap between a company's official mission statement and the raw data found in the "digital basement."

🛠️ How we built it

  • Orchestration: Built with LangChain and Google Gemini for high-reasoning risk detection.

  • Engine: Powered by Apify’s web data infrastructure. We utilize specialized Actors (Reddit Scraper, Glassdoor Scraper, and Google Search) to bypass bot detection and rate limits.

  • Intelligence: The agent uses an In-Memory Vector Store to index scraped data, allowing Gemini to cross-reference official claims against "Red Flag" signals.

  • Frontend: A clean, actionable dashboard built in Streamlit.

🚧 Challenges we ran into

Scraping high-value targets like Glassdoor and Reddit is notoriously difficult due to aggressive bot detection and dynamic layouts. We had to fine-tune our Apify Actor configurations to ensure we weren't just getting "noise" but high-signal technical reviews. Additionally, normalizing "Reddit slang" and "Legal jargon" into a single, cohesive Risk Dossier required complex prompt engineering to avoid hallucinations.

🏆 Accomplishments that we're proud of

We successfully built a functional MVP that can take a single company name and generate a multi-source "Risk Dossier" in under 5 hours. Being able to go from an idea to a working "Divergence Analysis" engine in 17 hours—leveraging the synergy between Claude, Gemini, and Apify—is a testament to the power of modern AI development.

🎓 What we learned

We learned that the most valuable data isn't always the most accessible. We discovered that "Digital Scent Trails"—small, repeated complaints on forums—are often more predictive of corporate failure than official audit reports. We also realized the importance of input validation to ensure the agent stays focused on legitimate business entities rather than generic web searches.

🚀 What's next for D3 · Deep Due Diligence

  • Direct Legal Integration: Moving beyond web results to link directly with the CourtListener REST API for real-time federal docket monitoring.

  • Visual Forensics: Using Gemini’s multimodal capabilities to analyze photos from social media for physical signs of equipment wear or site neglect.

  • Continuous Monitoring: An "Alert" system that notifies users the moment a high-divergence review or lawsuit is filed against a watched entity.

Built With

Share this project:

Updates