Fin Agent

💰 Fin Agent

AI-powered personal finance management with conversational insights

Home Lab Project · Multi-Bank Import · Financial Advisor Chat

The Problem

Personal finance is fragmented and hard to analyze:

Multiple accounts - Credit cards, savings, retirement, investments across institutions
No unified view - Each bank has its own app, its own categories
Shallow insights - Bank apps show spending, not strategy
No forecasting - What does my retirement look like? How does inflation affect me?

I wanted a single system that could import all my transactions, understand my finances holistically, and answer questions like a financial advisor.

The Solution

Fin Agent imports transactions from multiple banks, normalizes them into a unified format, and provides an AI-powered conversational interface for financial analysis.

"How much did I spend on groceries last quarter?"

"What's my projected retirement balance at 60?"

"How does my spending compare month over month?"

"What should I do with my savings right now?"

Key Features

📥 Multi-Source Import - Banks, retirement accounts, investment platforms
💬 AI Financial Advisor - Conversational insights via Valet LLM
📊 Spending Analysis - Category breakdown, trends, comparisons
🏦 Retirement Planning - Retirement account projections, withdrawal strategies
💸 Tax Optimization - Bracket calculations, deduction opportunities
🏠 Housing Analysis - Rent vs buy, mortgage scenarios
🛡️ Insurance Tracking - Health cover, MLS implications
📈 Inflation Modeling - Future expense projections

Architecture

flowchart TB
    subgraph Import["Data Import"]
        F[CSV/PDF Files]
        P[Parsers<br/>Bank, Retire, Invest]
    end
    
    subgraph Core["Finance Agent"]
        S[Storage<br/>Transactions]
        A[Analysis<br/>Spending, Trends]
        ADV[Advisor<br/>AI Chat]
    end
    
    subgraph Services["Specialized Services"]
        R[Retirement]
        T[Tax]
        H[Housing]
        I[Insurance]
        INF[Inflation]
    end
    
    subgraph Backend["Backend"]
        V[Valet Runtime<br/>LLM]
        K[Knowledge Base<br/>Local Finance Rules]
    end
    
    F --> P --> S
    S --> A --> ADV
    ADV --> R & T & H & I & INF
    ADV --> V
    ADV --> K

Supported Formats

Type	Format	Data
Credit Card	CSV	Transactions
Savings Account	CSV	Transactions
Retirement Accounts	CSV	Contributions, balance
Investments	CSV	Holdings, performance
Bank Statements	PDF	Coming soon

Parsers normalize transactions into a standard format with categories, merchants, and amounts.

The Advisor Chat

The core of Fin Agent is the conversational advisor:

flowchart LR
    subgraph Input["User Question"]
        Q[Query]
    end
    
    subgraph Detection["Intent Detection"]
        D{Keyword<br/>Analysis}
    end
    
    subgraph Data["Data Gathering"]
        S[Spending]
        R[Retirement]
        T[Tax]
        H[Housing]
    end
    
    subgraph Context["Context Assembly"]
        K[Knowledge Base]
        P[User Profile]
        TX[Transactions]
    end
    
    subgraph LLM["LLM Response"]
        V[Valet Runtime]
    end
    
    Q --> D
    D -->|spending keywords| S
    D -->|retirement keywords| R
    D -->|tax keywords| T
    D -->|housing keywords| H
    S & R & T & H --> Context
    Context --> V

The system detects what data is relevant to your question, gathers it from the appropriate services, and injects it into the LLM context.

Specialized Services

🏖️ Retirement Service

Retirement account projections
Withdrawal age calculations
Safe withdrawal rate analysis
Contribution optimization

💸 Tax Service

Tax bracket calculations
Deduction tracking
Estimated tax liability
Refund projections

🏠 Housing Service

Rent vs buy analysis
Mortgage scenarios
Stamp duty calculations
Affordability assessments

🛡️ Insurance Service

Health insurance tracking
Healthcare levy impact
Coverage analysis

📈 Inflation Service

Future expense modeling
Cost of living adjustments
Retirement purchasing power

Knowledge Base

The advisor is grounded in local financial knowledge:

Retirement account rules and withdrawal ages
Tax brackets and deductions
Healthcare levy thresholds
Investment principles

This prevents the LLM from hallucinating incorrect financial information.

API Endpoints

Category	Endpoints
Imports	Upload files, list imports
Accounts	List accounts, balances
Transactions	Search, filter, categorize
Analysis	Dashboard, spending breakdown
Advisor	Chat, sessions, history
Retirement	Projections, scenarios
Tax	Calculations, estimates
Housing	Scenarios, comparisons

Tech Stack

Component	Technology
API	FastAPI
Storage	JSON files (simple, portable)
LLM	Valet Runtime
Parsers	Custom per-institution
UI	SvelteKit dashboard

Privacy Note

This is a personal finance tool for my own use:

All data stays on my local servers
No cloud sync or external APIs (except Valet for LLM)
Transaction data never leaves my network
LLM queries don't include raw transaction details

What I Learned

Parsers are fragile - Each bank has its own CSV format quirks
Categories are hard - Merchant names don't map cleanly to categories
Context is key - The LLM needs real data to give useful advice
Knowledge grounding works - Injecting local finance rules prevents hallucinations

What's Next

Automatic categorization via LLM
PDF statement parsing
Goal tracking (savings targets, debt payoff)
Investment rebalancing suggestions

Fin Agent - AI Personal Finance Assistant