Project Story: Caldwell University Assistant Inspiration I wanted to create an AI-powered assistant to help Caldwell University students quickly access information such as event dates, professor contacts, and academic forms without needing to navigate through multiple pages manually.
What I Learned Building a Retrieval-Augmented Generation (RAG) system.
Processing and embedding PDF documents.
Using Pinecone for scalable vector storage.
Improving chatbot responses through prompt engineering.
Managing flexible search queries, including partial names.
How I Built It Extracted and cleaned text from university PDFs and spreadsheets.
Split content into manageable chunks for better embedding quality.
Embedded data using an embedding model and stored vectors in Pinecone with metadata.
Used Gemini 2.0 Flash as the language model for generating responses.
Designed prompts to produce structured, polite, and paragraph-separated answers.
(Optional) Built a React and Tailwind CSS frontend for a web-based chatbot interface.
Challenges Handling PDFs with inconsistent formatting and missing data.
Managing variations in name searches (full name vs. first name).
Ensuring that AI-generated answers stayed clear, polite, and user-friendly.
Maintaining error handling and robustness for API failures.
Tech Stack Backend: FastAPI (Python)
LLM: Gemini 2.0 Flash
Vector Database: Pinecone
Data Sources: PDFs, CSVs, Manual Entries
Frontend (Optional): React, Tailwind CSS
Log in or sign up for Devpost to join the conversation.