FinGraph

Understanding Freelancer Financial Stability

Comment

Introduction

On October 31st, 2023, Bloomberg reported that independent contractors and freelancers are increasingly essential—yet still sidelined by traditional finance:
“Freelancer reliance rises in U.S., with 20% of corporate work now done by them.”
Source: Bloomberg

Despite earning steadily, many freelancers continue to be flagged as “high-risk” simply because their income does not resemble W-2 employment.

A 2022 McKinsey report reinforced the scale of this shift:
36% of U.S. employed adults—58 million people—identify as independent workers.
Source: McKinsey & Company

Yet nearly 49% of gig and 1099 workers have been denied financial services they could afford, largely due to irregular or unverifiable income.
Source: Business Wire

Freelancers are financially active but remain structurally invisible. Their money flows across clients, subscriptions, transfers, bills, and merchants—patterns that behave like a graph, not a paycheck.

Traditional credit systems flatten this richness into a single number.
FinGraph exists to address this gap.

Inspiration, Problem and Why FinGraph ?

The idea originated from hearing the same story repeated across design, consulting, gig work, and creator communities:
“My income is stable, but the system still rejects me.”

This happens because legacy credit models expect:

  • Single-employer income
  • Predictable pay cycles
  • Low spending variability
  • Traditional bills and obligations

Freelancers violate these assumptions. Their financial behavior includes:

  • Multi-sourced, asynchronous income
  • Irregular bills and subscriptions
  • Client concentration risk
  • Month-to-month cashflow shifts

The issue isn’t poor financial behavior—it’s a model built for an outdated workforce.

FinGraph evaluates freelancers based on real behavior, not rigid templates.

Instead of asking “How much did you make?”, FinGraph asks:

  • How stable is your income?
  • How diversified are your clients?
  • Which subscriptions drive expenses?
  • How volatile is your cashflow?
  • How much savings runway do you have?

These are graph questions.

How it works ?

1. Data Layer — Real Banking Data

FinGraph reconstructs a freelancer’s financial world using the Capital One Nessie API
→ http://api.nessieisreal.com/

All data—accounts, deposits, withdrawals, transfers, merchants, and bills—is modeled as nodes and relationships inside Neo4j AuraDB.

2. Intelligence Layer — Multi-Agent Reasoning (ADK)

On top of the graph, FinGraph runs a structured multi-agent pipeline using Google’s Agent Development Kit (ADK). The agents are:

  • Income Stability Agent – income cycles and irregularity
  • Volatility Modeling Agent – cashflow variance and drawdowns
  • Reliability Agent – client concentration, late payments, subscriptions
  • Risk Scoring Agent – Financial Identity Score + savings runway
  • Explainer Agent – grounded narrative from structured metrics

Agents communicate through JSON-based state passing, ensuring the reasoning remains grounded in real data and avoids hallucinations.

3. Application Layer — Dashboard

  • Backend: Python + FastAPI + ADK
  • Frontend: React + Neovis.js
  • Deployment: Cloud Run

The dashboard presents:

  • The Financial Identity Score
  • Component breakdowns
  • Timelines for income and cashflow
  • Client concentration insights
  • Savings runway
  • An interactive Neo4j financial relationship graph
  • Human-readable recommendations

Challenges we encountered

  • Designing a unified graph schema from heterogeneous API data
  • Writing precise Cypher queries to support our tools
  • Constructing a multi-agent ADK pipeline with reliable state sharing
  • Preventing LLM hallucinations
  • Integrating ADK, Neo4j, FastAPI, and React cohesively
  • Rendering large graph structures without overwhelming users

Every part required careful debugging and alignment across data, agents, and UI.

Conclusion

FinGraph reframes financial stability for a world increasingly powered by independent work. By combining graph databases, real banking data, and Gemini-powered multi-agent reasoning, it transforms fragmented transactions into a clear, human-centered Financial Identity. This isn’t a budgeting tool—it’s a new interpretation of financial identity itself, one that tells a story traditional credit scores never could.

Our approach naturally aligns with the goals of Technica, Intuit, Capital One, and MLH: we use AI responsibly, apply real financial data to solve an underserved problem, promote transparency and fairness, and build inclusive technology that empowers freelancers.

Accomplishments

  • Fully working multi-agent reasoning pipeline
  • Real financial graph from Nessie API ( ~87,000 accounts processed )
  • Transparent scoring model
  • Clean, fast dashboard ( < 4 seconds)
  • Integrated explainability across all steps

What we learned

We learned to design graph-based financial models, orchestrate multi-agent LLM systems, generate grounded explanations, and build real-time analytics pipelines.

What's next for FinGraph

  • Forecasting income and expenses
  • Scenario simulation (“What if my top client leaves?”)
  • Multi-bank support
  • Personalized guidance beyond scoring
  • Adaptive user insights

Testing Credentials

For evaluation and demonstration purposes, the following sample credentials have been provided.

Credential 1: Username: [email protected] Password: f20dd5184160

Credential 2: Username: electronic [email protected] Password: 35571461272f

Please note that these credentials are intended only for testing in a controlled environment and do not provide access to any production-level or sensitive information.

Built With

Share this project:

Updates