FinGoal – Agentic Budgeting, Reimagined

Home Page
Chat interface
Goals dashboard

💡 Inspiration

Every budgeting app treats you like a stranger. You open it, dump in your numbers, get a generic pie chart, and close it only to repeat the whole process next week because it forgot everything about you.

I wanted to build something different: a financial assistant that actually remembers you. One that understands you're saving for a house, knows your rent went up last month, and can say "hey, based on what you told me last week, here's how that changes your timeline" without you re-typing your entire financial history every single time.

That's how FinGoal was born: an Agentic AI budget assistant that treats personal finance the way a real financial advisor would as an ongoing relationship, not a one-off transaction.

🚀 What it does

FinGoal is a conversational financial assistant that:

Understands your goals in plain English — tell it "I want to save ₹5L for a car in 18 months" and it builds a real savings plan around it
Remembers everything — thanks to persistent conversational and user memory, it maintains context across sessions instead of starting from zero
Analyzes your finances automatically — upload a statement, and it extracts insights, flags overspending, and suggests optimizations
Adapts in real time — every conversation updates your goals, budget, and dashboard automatically, so your financial picture is always current
Visualizes progress — an interactive dashboard turns raw numbers into a clear, trackable path toward your goals

🛠️ How I built it

FinGoal is built as a layered, containerized system:

``` React + TypeScript Frontend │ ▼ FastAPI Backend │ ▼ Conversation & Goal Engine ──▶ Financial Planning Engine │ ▼ Snapshot Generation │ ▼ PostgreSQL Database ```

Frontend (React + TypeScript): a responsive, interactive dashboard that reflects the user's live financial state
Backend (FastAPI): orchestrates conversation flow, goal logic, and financial computations
Conversation & Goal Engine: the "agentic" core — interprets intent, updates existing goals instead of duplicating them, and drives follow-up questions based on prior context
Financial Planning Engine: runs the actual budgeting math — savings timelines, expense optimization, affordability checks
Snapshot Generation: periodically captures the user's financial state so the dashboard reflects the latest conversation instantly
PostgreSQL: persists users, goals, conversations, and financial snapshots
Docker Compose: one command (`docker compose up --build`) spins up the entire stack — no manual environment setup

🧗 Challenges I ran into

Making memory useful, not just present. Storing conversation history is easy; deciding what matters — a changed goal, an updated income, a new expense — and folding it back into the planning engine without contradicting earlier context was the hardest part.
Keeping the agent from re-asking what it already knows. I had to carefully design the goal-update logic so the assistant edits existing goals instead of creating duplicates every time the user's situation shifts.
Syncing conversation state with the dashboard in real time, so what the user says in chat is immediately reflected visually — without manual refreshes or stale data.
Parsing messy, real-world financial statements into structured data the planning engine could actually reason about.

🎓 What I learned

How to design agentic systems that persist state meaningfully across sessions, not just log transcripts
The nuance in building conversational memory that updates and corrects itself rather than just accumulating
How to bridge an LLM-driven conversational layer with a deterministic financial planning engine so the numbers stay accurate even when the conversation is fuzzy
The value of a clean, Dockerized architecture for fast iteration across a full-stack, multi-service project