Heritage Square Foundation

Accessing your drive files with ease..

Comment 3

What it does:

Our team developed an AI agent with Google Drive using Python, Langchain and RAG to assist in document organization, tagging, and retrieval. Implementing AI-driven document categorization, suggesting reorganization strategies, for enhanced prompting capabilities. The nonprofit organization we developed for is Heritage Square Foundation.

The tools/languages we used:

Python, LangChain, RAG, Google Cloud API, Embedding & Vector Store in FAISS, React, Vite Framework

Challenges We Ran Into

  • Time Zone Differences: Coordinating meetings across four different time zones proved difficult, limiting opportunities for real-time collaboration.
  • Learning Curve with New Tools: Implementing Retrieval-Augmented Generation (RAG) using LangChain and integrating Gemini Pro required rapid learning and experimentation with unfamiliar frameworks and APIs.
  • Service Account Limitations: Google service accounts lacked full access to Drive contents unless folders were manually shared. This restricted our ability to fully deploy the file organization and cleanup features.
  • Categorization Accuracy: Automatically classifying mixed-content files (e.g., PDFs with event data and donor info) posed challenges, occasionally resulting in misplacements.
  • AI Response Handling: Some error messages were mistakenly returned as AI responses, which affected user experience and required additional filtering logic.
  • Connectivity & Power Issues: Team members experienced internet instability and power outages during critical development and integration stages.
  • Merge Conflicts & Deployment Stress: Coordinating last-minute code merges and resolving deployment issues under time pressure tested our version control and debugging skills.

Accomplishments We're Proud Of

  • Built Our First AI Agent: As a team completely new to AI agent development, we successfully designed and implemented an intelligent assistant using RAG (Retrieval-Augmented Generation) and LangChain, a major leap in our technical growth.
  • Integrated Gemini Pro (Text + Vision): We effectively used Gemini Pro to analyze both text and images for intelligent document understanding and classification.
  • End-to-End Full-Stack Development: From the backend (FastAPI, Python) to the frontend (React, Vite), we built a fully functioning application capable of interacting with Google Drive using natural language.
  • Authenticated and Queried Google Drive via API: We successfully authenticated users with OAuth and enabled secure document retrieval, movement, and cleanup within Google Drive.
  • Created a Scalable Architecture: Our modular system supports future expansion, including additional file types, smarter search capabilities, and potential Gmail/Sheets integration.
  • Collaborated Across Time Zones: Despite working across four time zones and facing connectivity issues, we maintained momentum, met deadlines, and delivered a solution that works.

What We Learned

Through this project, we deepened our understanding of integrating AI with real-world workflows. We gained hands-on experience in:

  • Leveraging Gemini Pro for both text and image analysis
  • Structuring unorganized cloud storage using LLMs and embeddings
  • Building full-stack applications with React and FastAPI
  • Collaborating across distributed teams under tight deadlines

What's Next for Heritage Square Foundation

To ensure long-term impact, our next steps include:

  • Finalizing deployment using service accounts
  • Improving file categorization accuracy and handling edge cases
  • Expanding capabilities to support Google Sheets, Forms, and Gmail
  • Training Heritage Square staff to confidently use and maintain the assistant

Built With

Share this project:

Updates