AutoFi – Agentic AI Financial Assistant

Applied for Startup Track, NVIDIA Challenge, and CapitalOne Challenge

Inspiration

We were inspired by the millions of Americans living in banking deserts—communities where the nearest bank branch is miles away, if one exists at all. For these individuals, managing money means driving 45 minutes to cash a paycheck and paying 3–5% in fees at check-cashing stores, sending money to family through predatory wire services that take $15 from every $100, and facing rejection for basic financial services because they lack a credit history that traditional banks recognize.

Without access to savings accounts, they hide cash in their homes, vulnerable to theft and unable to build interest. Without checking accounts, they purchase money orders to pay rent and utilities, spending $1.50 per transaction on services most people take for granted. When emergencies strike—a medical bill, a broken-down car, an unexpected layoff—they turn to payday lenders charging ridiculously high APRs because no traditional institution will help them, trapping them in cycles of debt that can take years to escape.

We wanted to do our part in helping this issue by creating an agentic AI financial assistant that provides intelligent, automated financial management accessible to everyone, regardless of their location or banking access.

What It Does

AutoFi is an agentic AI platform that helps users manage and optimize their finances in real time. It analyzes spending data, identifies inefficiencies, and automates key financial actions with user approval.

Multi-Agent Workflow

Orchestrator Agent Banking Agent Budgeting Agent Trading Agent Stock Anaylsis Agent Research Agent Email Agent Response Agent

Core Features

Subscription Management:

Detects recurring charges and lists all active subscriptions.
Drafts cancellation emails automatically and sends them after user confirmation.
Runs through a full DocGen → Response → Email reasoning sequence for transparency.

Smart Budgeting & Insights:

Analyzes overspending and reallocates savings dynamically.
Displays real-time insights such as “Saved $55 by adjusting entertainment budget.”

Investment & Allocation (Prototype):

Uses Alpaca Paper Trading API to simulate ETF and stock purchasing logic.
Demonstrates agent-based portfolio recommendations under user control.

Conversational Interface:

Visualizes each agent’s “thinking” process step-by-step.
Designed to feel like a chat with a transparent, human-like financial advisor.

How We Built It

Frontend (React Native + Expo):

Built a cross-platform mobile interface with Expo, NativeWind, and Reanimated.
Designed themed UI components and gradient-based visuals for clarity and trust.
Added dynamic animations for agent transitions and “thinking” sequences.

Backend & AI System:

Implemented a multi-agent reasoning architecture modeled after LangChain.
Each agent performs distinct tasks (data parsing, insight generation, action drafting).
Leveraged Nemotron for reasoning, summarization, and email generation.
Integrated explicit approval checkpoints for every financial action.

Integrations:

Plaid API – Transaction aggregation and subscription detection.
Stripe Financial Connections – Secure account verification and balance retrieval.
Alpaca API – Simulated investment execution.
Expo MailComposer – Draft and preview cancellation emails directly in-app.

Security:

Encrypted data handling and secure token storage.
Role-based permissions for agent actions.
Full audit logs for AI reasoning and user approvals.

Challenges We Faced

Building Trust: Users hesitate to let AI handle finances. We solved this by showing each agent’s reasoning process visually.
Data Standardization: Each financial API returned data differently; we had to normalize structures before analysis.
Limited APIs: Many companies lack cancellation APIs, so AutoFi generates natural-language emails for user-approved execution.
Development Speed: Coordinating UI, agents, and financial APIs within a short timeframe required careful modularization.

Accomplishments That We’re Proud Of

Built a fully functional multi-agent pipeline for analyzing, reasoning, and acting on financial data.
Designed a transparent, interactive UI that visualizes how each AI decision is made.
Enabled end-to-end functionality:

Analyze transactions
Identify subscriptions
Draft cancellation requests
Confirm user approval
Execute actions safely
- Early results show users can recover $100–$200 per month through AutoFi’s optimization features.

What We Learned

How to design agentic architectures that balance autonomy with accountability.
That transparency builds user trust more effectively than pure automation.
How to merge financial APIs, AI models, and real-time mobile UX into a cohesive product.
The importance of gradual automation — always keeping the user in the decision loop.

What’s Next for AutoFi

Expanded Agent Chaining: More advanced reasoning between financial planning and investment modules.
Automated Bill Negotiation: Agents that contact providers to reduce rates or cancel unused plans.
Adaptive Investment Advisor: Risk-based ETF recommendations using live market data.
Web Dashboard: Browser view for analytics, insights, and action approvals.
Accessibility Expansion: Multi-language support and simplified onboarding.
Regulatory Compliance: Aligning with Open Banking and AI safety standards.

Would you like me to also create a shorter Devpost summary version (around 250–300 words) that fits into the “Project Description” box? That’s often required separately and needs to hit only the key points.