Nixora

Nixora is an Autonomous Loan Intelligence for the Action Era

Comment

🧠 Nixora — Autonomous Loan Intelligence for the Action Era

Turning legal contracts into self-updating financial systems.

🤖 Gemini Integration

Nixora is architected entirely around the Gemini 3 model family, leveraging its advanced reasoning capabilities to transform static legal documents into active, self-managing financial objects. The integration is central to the application's core functionality in two distinct ways:

  1. Gemini 3 Flash for Structural Reasoning (The Engine) We utilize gemini-3-flash-preview configured with ThinkingLevel.LOW to power our LMA Extraction Engine. This feature is the backbone of the application. Instead of simple OCR, Gemini 3 Flash performs clause-level reasoning to map non-standardized, multi-hundred-page credit agreements into a strict, LMA-aligned JSON schema. The "low" thinking setting allows us to achieve high-fidelity data extraction with the low latency required for a seamless user upload experience.

  2. Gemini 3 Pro for Deep Financial Analysis (The Brain) We deploy gemini-3-pro-preview configured with ThinkingLevel.HIGH for the AI Loan Chat Assistant. This feature transforms the user experience from passive reading to active interrogation. We rely on Gemini 3 Pro's deep reasoning capabilities to understand complex cross-clause dependencies—such as how a specific "Material Adverse Effect" definition triggers a "Default Event." This allows users to ask high-stakes financial questions and receive answers grounded in legal logic rather than statistical probability.

By using the Google Gen AI Java SDK, we dynamically switch between these reasoning levels, making Gemini 3 the operating system of the entire loan lifecycle.

🚀 Inspiration

The global loan market runs on trillions of dollars of obligations, yet the infrastructure behind it still runs on PDFs, emails, and spreadsheets.

Every commercial loan is a long-lived system:

  • It evolves over years
  • It creates obligations every month
  • It carries risk that changes over time
  • It involves many parties, not just one borrower

But today, loans are treated as static documents, not living financial entities.

Nixora was built to change that.

We asked a simple question:

What if a loan could think, reason, and manage itself?

🧩 The Problem

Loan agreements are not simple files — they are multi-year programs written in legal language.

Inside every credit agreement are:

  • Pricing formulas
  • Conditional obligations
  • Timelines
  • Risk thresholds
  • Enforcement triggers

Yet institutions still manage them manually:

  • Analysts read documents
  • Lawyers interpret clauses
  • Operations teams track dates in spreadsheets
  • Compliance is done by human memory

This leads to:

  • Missed covenants
  • Late payments
  • Undetected risk
  • Expensive operational overhead

The financial system is digital — but loan intelligence is still human-powered.

🧠 Our Solution

Nixora turns loan agreements into autonomous digital agents.

Using Google Document AI, Vertex AI, and Gemini 3.0 Pro, Nixora converts raw loan contracts into:

Self-aware, continuously reasoning financial objects.

Each loan becomes:

  • A structured digital model
  • A reasoning engine
  • A task scheduler
  • A live risk monitor

Instead of users searching documents, Nixora’s Gemini-powered agents understand, track, and reason over the loan lifecycle.

🛠 How It Works

When a user uploads a loan:

  1. Document AI extracts raw text and tables
  2. gemini-3-flash-preview performs clause-level and cross-clause reasoning
  3. Vertex AI maps the result into an LMA-aligned loan model
  4. The loan becomes a live digital twin stored as structured JSON

Gemini now powers:

  • AI loan chat
  • Risk detection
  • Obligation tracking
  • Portfolio intelligence
  • Workflow automation

What it does

Nixora is an end-to-end Digital Loan Operating System for the global loan market.

Users can upload any loan agreement (PDF, Word, Excel), or capture documents directly from the web using the Nixora Chrome Extension. From there, Nixora transforms unstructured legal contracts into live, intelligent, and standardized loan data.

Nixora then:

  1. Extracts and normalizes text using Google Document AI
  2. Understands and structures the contract using Vertex AI + Gemini 3.0 Flash Preview
  3. Maps everything into LMA-aligned JSON — creating a real-time, structured loan model

From a single upload, Nixora powers the entire loan lifecycle:

  • Portfolio Search & Querying Instantly search across borrowers, lenders, facilities, covenants, currencies, margins, and maturities across your entire book of loans.

  • Loan Comparison Compare pricing, structure, repayment terms, and risk across multiple facilities and borrowers in seconds.

  • AI Loan Chat Ask natural language questions about any agreement — Gemini returns precise, explainable answers grounded directly in the contract.

  • Risk, Covenant & Maturity Reporting Automatically detect high-risk loans, upcoming deadlines, covenant exposure, and compliance gaps.

  • PDF & Excel Exports Generate institution-grade reports for credit committees, auditors, and regulators.

  • Loan Schedules & Event Tracking Automatically create payment schedules, covenant tests, and maturity timelines from the legal text.

  • Calendar & Workflow Sync Push obligations directly into Google Calendar and Trello to ensure nothing is missed.

  • Real-time Collaboration Invite colleagues via WhatsApp to view, edit, and work on the same loan live.

  • Push Notifications Desktop alerts for uploads, payments, covenant breaches, and critical risk events.

  • Chrome Extension for Live Loan Capture Capture loan documents, term sheets, and deal data directly from emails, data rooms, and web portals — sending them instantly into Nixora for structuring and tracking.

Nixora turns static loan documents into living financial systems that teams can search, monitor, collaborate on, and trust.

🧠 Gemini’s Role

Gemini 3 Pro acts as a financial reasoning engine that:

  • Reads entire multi-hundred-page agreements
  • Extracts legal meaning, not just text
  • Maintains state across time
  • Triggers tools and workflows

This directly matches the Gemini 3 “Action Era” vision: AI that plans, reasons, and executes.

📡 API Surface

🔐 Authentication & Security

  • POST /api/v1/auth/register
  • POST /api/v1/auth/login
  • POST /api/v1/auth/refresh
  • POST /api/v1/auth/logout

Forgot Password

  • POST /api/v1/forgot-password/verify-email/{email}
  • POST /api/v1/forgot-password/verify-otp/{otp}/{email}
  • POST /api/v1/forgot-password/change-password/{email}

Google OAuth

  • GET /api/google/connect
  • GET /api/google/callback

📄 Core (Loan Document Management)

  • POST /api/loans/documents/upload
  • PATCH /api/loans/documents/update-field/{loanId}
  • GET /api/loans/documents/getLoan/{loanId}
  • GET /api/loans/documents/getAllLoan
  • DELETE /api/loans/documents/delete-loan/{loanId}

📊 Portfolio Intelligence

  • POST /api/loans/query/search-query
  • POST /api/loans/compare
  • GET /api/reports/portfolio
  • GET /api/reports/high-risk
  • GET /api/reports/maturity

🤖 AI Loan Chat

  • POST /api/loan-chat/chat/{loanId}

📤 Reports & Exports

  • GET /api/reports/export/portfolio/pdf
  • GET /api/reports/export/portfolio/excel
  • GET /api/reports/export/maturity/pdf
  • GET /api/reports/export/maturity/excel
  • GET /api/reports/export/high-risk/pdf
  • GET /api/reports/export/high-risk/excel

🗓 Scheduling & Workflow

  • GET /api/loans/{loanId}/schedule
  • POST /api/loan/calendar/{loanId}/sync

🔗 Integrations (Trello)

  • POST /api/trello/connect
  • POST /api/trello/callback
  • GET /api/trello/boards
  • GET /api/trello/boards/{boardId}/lists
  • POST /api/trello/select-list
  • POST /api/trello/push-field

👥 Collaboration

  • POST /api/collaboration/invite/{loanId}
  • POST /api/collaboration/accept

🔔 Notifications

  • POST /api/notifications/register-device

🧩 Architecture

Nixora System Architecture

⚙️ Challenges We Faced

  • Accurate legal understanding Regex and template matching failed on complex loan language. Gemini 3.0 Pro reasoning across clauses was required to correctly interpret obligations and exceptions.

  • LMA standardization There is no open canonical LMA schema. We reverse-engineered one from public agreements and industry references to create a realistic digital loan model.

  • Real-world data Commercial loans are confidential. We curated and validated against real publicly available agreements to ensure accuracy.

  • Integrating Gemini 3.0 Pro with Java and Gen AI SDK We encountered significant challenges integrating the Gemini 3 Pro Preview using the Java Gen AI SDK, specifically regarding dependency conflicts and "Publisher Model Not Found" errors due to regional endpoint limitations. Through rigorous debugging, we successfully migrated our architecture to the global endpoint and implemented the correct ThinkingLevel enum configurations, establishing a stable connection to the Gemini 3 API.

🏆 Accomplishments

  • Fully digitized, live loan objects
  • LMA-aligned structured contract models
  • Gemini-powered loan chat & reasoning
  • Automated schedules & risk detection
  • Calendar, Trello & notification integrations
  • Institutional-grade reports & exports

📚 What We Learned

  • Financial professionals don’t want more dashboards — they want their existing tools to become intelligent
  • AI becomes exponentially more powerful when combined with structured financial data
  • Automation beats analytics for real operational impact

🚀 What’s Next

With Gemini at its core, Nixora is expanding into:

  • Agentic Document Extraction (ADE)
  • Autonomous covenant enforcement
  • Portfolio-level risk agents
  • Secondary-market trading intelligence
  • Predictive default detection

Nixora is becoming the AI operating system of global Loan credit.

Built With

Share this project:

Updates