TrustPort

TrustPort is an AI Credit Passport for immigrants. We turn foreign banking data into a trusted "Shadow Score" and landlord verification letter, making the credit invisible, visible.

Comment 40

💡 Inspiration

Millions of skilled professionals move abroad every year—engineers, doctors, and students—only to find that their financial reputation resets to zero. They are "Credit Invisible."

I was inspired by the story of a certain someone with perfect financial discipline in his home country who couldn't rent an apartment in his new city because he had no local credit score. I realized that in a globalized world, credit data is still stuck within borders. I built TrustPort to be the bridge.

🚀 What it does

TrustPort is a Global Credit Passport that translates foreign banking history into local trust. It uses a multi-model AI pipeline to analyze raw documents and generate a verified "Shadow Credit Score."

  1. AI Fraud Detection: Before analyzing finances, I use computer vision to scan uploaded bank statements for pixelation, font inconsistencies, and tampering.
  2. Solvency Analysis: I extract transaction data to calculate monthly income, savings rates, and volatility, generating a "Shadow Credit Score" (300-850) based on Basel III and IFRS 9 risk concepts.
  3. Visual Trust: The Flutter dashboard visualizes cash flow trajectories and risk flags using a banking-grade design system.
  4. Verification Letter: The system generates a downloadable, watermarked PDF letter for landlords, explaining the applicant's financial health in plain English.

⚙️ How we built it

I engineered a Hybrid AI Ensemble to leverage the specific strengths of two different state-of-the-art models.

The "Hybrid" Backend (Python/FastAPI on Render)

I built a robust backend using FastAPI hosted on Render. The core innovation is our two-step AI pipeline:

  • The "Eyes" (Google Gemini 2.0 Flash): We route the raw document image to Gemini Vision to perform OCR and a Fraud Check. It inspects the document for visual artifacts that indicate Photoshopping or tampering.
  • The "Brain" (OpenAI GPT-4o): Once the text is sanitized of PII (Emails/SSNs redacted via Regex), we send it to GPT-4o. I engineered a system prompt based on FATF (Anti-Money Laundering) and Basel III credit risk principles to categorize transactions and predict future cash flow.

The "Native" Frontend (Flutter Web on Netlify)

I initially prototyped in Streamlit but pivoted to Flutter Web to deliver a pixel-perfect, native-app experience.

  • Dynamic Visualization: I used fl_chart for the predictive cash flow graph and syncfusion_flutter_gauges for the reactive credit meter (which glows Teal or Red based on risk).
  • Client-Side PDF Engine: The app generates the verification letter and applies the TrustPort watermark directly in the browser using the Dart pdf package, ensuring the document is created instantly without server latency.

🚧 Challenges we ran into

  • The UI Pivot: Switching from Streamlit to Flutter mid-hackathon was a massive risk. I had to rebuild our layout logic from scratch, but it paid off with a much smoother "Fintech" aesthetic.
  • Hallucinations vs. Math: LLMs are great at words but can be bad at math. I had to implement a hybrid parsing logic—using Python (pdfplumber) for raw number extraction and AI only for the classification and reasoning, ensuring our "Shadow Score" was mathematically accurate.
  • Handling PDF Fraud: I realized that text extraction alone wasn't enough; anyone can edit a PDF text layer. Integrating Gemini Vision to "look" at the document image solved this, adding a layer of security that text-only parsers miss.

🏆 Accomplishments that we're proud of

  • The "Vision Shield": Successfully implementing Gemini to detect document tampering gave our app a security edge that most hackathon projects lack.
  • The "Shadow Score" Algorithm: I successfully built a logic model that penalizes volatility and rewards savings consistency, creating a fair metric for credit-invisible users.
  • Real-time PDF Generation: Generating a watermarked, professional landlord letter client-side in Flutter feels magical and provides immediate value to the user.

🔮 What's next for TrustPort

  • Real Open Banking: Integrating Plaid or Teller APIs to fetch live bank data instead of PDF uploads.
  • Landlord Portal: A view specifically for landlords to receive verified "TrustPort Links" directly from applicants.
  • Blockchain Identity: Storing the "Shadow Score" on-chain so users own their reputation permanently.

Built With

+ 47 more
Share this project:

Updates