Inspiration
The dream of homeownership is universal, yet the process of achieving it often remains frustratingly complex. Aspiring buyers frequently struggle to formulate a concrete, actionable financial strategy.
Our team was inspired to bridge this gap by replacing outdated static financial calculators with a dynamic, engaging, and highly personalized conversational experience. We aimed to transform confusing financial planning into a tangible roadmap.
To make the often intimidating topic of personal finance feel immediately accessible and fun, we chose a pixel art aesthetic, drawing visual inspiration from the charming dialogue systems of retro RPGs and adventure games. This approach turns a daunting task into an enjoyable quest guided by our friendly AI coach, Homy.
What it does
Home Sweet Home acts as Homy, an AI financial coach presented in a charming, nostalgic pixel art interface. Users can chat naturally with Homy about their homeownership goals (e.g., "I want to save $500 a month" or "I need to buy a house in five years").
Conversational Guidance: Homy handles the full dialogue, extracting user data (income, savings, target time/price).
Dynamic Problem Solving: The system instantly processes user intent and solves the one missing financial variable using one of three Solver Modes (calculating the required time, budget, or monthly savings).
Engaging UI: Features a fully animated character that reacts with different emotions and a retro text-box that uses a click-through typing effect for immersion.
How we built it
We built the system using a strict, three-layer architecture designed for stability, scalability, and accurate financial computations.
Frontend (UI/UX)
- Implemented entirely with pure HTML/CSS and JavaScript to maintain the low-fi, pixel-perfect retro aesthetic.
- JavaScript manages all persistent chat state, the interactive click-through typewriter effect, and dynamically swaps character sprites based on emotion tags returned by the backend.
AI Service (API Layer)
- A FastAPI backend acts as the single orchestration point.
- It receives both the latest user message and the full chat history.
- The service uses the OpenAI API with strict Function Calling, forcing the LLM to output structured data that matches a Pydantic schema (
CalculationRequest). - This ensures Homy only calls the solver when all required variables for one of the three solver modes (time, budget, savings) are properly collected.
Logic Engine (Solver)
- A standalone Python module (
financeengine.py) performs all financial calculations (compound interest, DTI validation, closing-cost estimation). - It receives clean, structured JSON directly from the Function Call.
- This clean separation guarantees numerical correctness independent of the conversational layer.
Challenges we ran into
Bridging Natural Language and Strict Logic
Translating vague human language (e.g., “I want something soon in a big city”) into a precise, multi-field JSON schema was the hardest part. The biggest difficulty was teaching the LLM to leave exactly one field as null, because that field is the one the solver must compute.
Managing Asynchronous State
Ensuring the UI remained fully locked during the entire backend sequence (API call → LLM decision → solver invocation → LLM final message) required careful JavaScript state management. Without it, users could accidentally fire multiple backend requests and break the chain.
Pixel-Perfect Responsiveness
Achieving a crisp retro feel demanded meticulous CSS tuning. We used image-rendering: pixelated combined with fluid viewport units (vw/vh). Small layout changes risked breaking sprite placement or causing overlap with the chat box on different screen sizes.
Accomplishments that we're proud of
Robust Function Calling
We successfully built a complex multi-step tool-use system where the LLM gradually extracts user intent, assembles a complete data structure, calls a Python solver, and then interprets the results in character as Homy.
Fully Integrated Experience
The emotional reactions of the character (e.g., [CELEBRATE]) are tied directly to numerical outcomes from the backend model, creating a uniquely synchronized blend of creative design and technical accuracy.
Complete MVP in 36 Hours
We delivered a full-stack product featuring a pixel-art front end, a resilient API layer, a database-free solver, and a complete LLM wrapper — all built, debugged, and integrated within the hackathon’s timeframe.
What we learned
We gained a deep understanding of how to design resilient agentic workflows.
One of the biggest takeaways: extremely strict Pydantic schemas combined with detailed system prompts outperform short or fuzzy instruction prompts by a huge margin.
By defining exact rules for when and how the AI should use tools, we dramatically increased consistency and reduced hallucinations.
What's next for Home Sweet Home
Authentication & Persistence
Implement user accounts and database-backed storage (e.g., PostgreSQL or MongoDB) so users can save their “Adventure Progress” and financial profiles. This would allow Homy to remember previous conversations, evolving savings plans, and updated life circumstances, turning a single session into a long-term coaching journey.
Deeper Scenario Simulation
Extend the solver to handle richer what-if scenarios (interest-rate shocks, income changes, different regions or property types) and let users compare multiple timelines side by side as different “quests.”
Built With
- css
- fastapi
- html
- javascript
- openai-gpt-4o-mini
- pydantic
- python