-
Closing.wtf hero
-
-
How closing.wtf works
-
Loan Estimate Summary
-
Summary of Closing Costs by Negotiable or Non-Negotiable
Inspiration
Loan estimates are complicated and include many hidden closing costs and potentially unfair rates. Homebuyers often get screwed overpaying for negotiable services and not knowing a seller is responsible for certain closing costs.
What it does
Closing.wtf analyzes real estate closing documents and cross references the details with an extensive knowledge graph of real estate standards across the USA to find mistakes, potential fraud, and help the buyer negotiate the best deal.
How we built it
User and Expert Interviews: Initiated our design process by interviewing target users (home buyers) to identify pain points and gain insights. Data Management: Imported real estate disclosure data into a vector database. Information Extraction: Employed LlamaParser to extract clean information from loan estimate PDFs, ensuring data quality and accessibility. Framework Integration: Used LlamaIndex to seamlessly integrate GPT-4, forming the backbone of our system's architecture. Prompt Engineering: Enhanced both the quality of input and output through meticulous prompt engineering in the backend, optimizing user interactions.
Challenges we ran into
Building a Knowledge Graph: Mortgages can involve many parties and have complex loan structures, requiring extensive schemas. Desire for Structured Outputs: Transitioned from conversational to structured outputs via LLM or backend processing to provide users with clear, intuitive property disclosures. Instead of conversation style outputs, we want structured outputs (by either LLM or backend processing). So that users receive clean data on their interested properties, and have clean, intuitive disclosure analysis. Prompt Engineering: Invested in prompt engineering to ensure clean, formatted data presentation.
Accomplishments that we're proud of
Simple User-Centric Design: Prioritized user needs by empathetically addressing pain points through thoughtfully designed features. Effective Technology Use: Leveraged RAG and LlamaParser for efficient data extraction and storage, ensuring clean and structured outputs. Business Potential: Recognized the significant commercial opportunity our project presents.
Originality: Inspired by our real-life experiences, our project stands out due to its uniqueness and relevancy.
What's next for Closing.wtf
Better Recommendations: Improve our knowledge graph to give better recommendations on how users can detect mistakes and get better value from their mortgage.
Other Document Integrations: Implement other document types such as the closing documents to detect mistakes from lenders, other parties, and title companies.
Log in or sign up for Devpost to join the conversation.