DISCLAIMER: This chatbot is an independent learning project developed solely for educational purposes. It is not an official resource of Bartlesville High School or any Bartlesville Public School entity. While responses are generated from the Bartlesville High School Student Handbook, the accuracy, completeness, or currency of information provided cannot be guaranteed. Users should always verify information directly with official school resources before making decisions or taking action based on the chatbot’s responses. This tool is hosted using free and open-source resources; therefore, availability, performance, and response times may vary.
About the Handbook Chatbot:
The Handbook Chatbot represents my first step beyond classroom exercises into creating a practical, real-world AI application. Designed around the Bartlesville High School student handbook, it serves as a valuable learning sandbox to explore and deepen my understanding of Large Language Models (LLMs). Through iterative development, this chatbot allows me to continuously sharpen my skills, experiment with new concepts, and expand my technical expertise in NLP, retrieval techniques, and application development—all while providing a useful resource to quickly access handbook information.
This AI-powered chatbot that provides quick access to information from the Bartlesville High School student handbook. The system features a FastAPI backend for processing queries, a Streamlit frontend for an intuitive user experience, and ChromaDB for efficient document retrieval.
Key Features:
- Natural Language Processing (NLP): Uses LLM (Llama) to generate responses based on the handbook.
- Document Processing & Vector Search: Segments the handbook into semantic chunks, stored and retrieved using ChromaDB for better search accuracy.
- Query Validation & Filtering: Enhances question clarity and filters inappropriate content.
- FastAPI Backend: Ensures efficient processing of chatbot interactions.
- Streamlit Frontend: Provides an easy-to-use web interface for students and faculty to quickly find information.
- Application Link: Handbook Chatbot
Roadmap
To see a list of things I want to learn that will also translate into new features, capabilities and improvements to the chatbot. See Chatbot Roadmap.
2025-03-30: Version 2.0.1
- Updated security update to prevent bots and nefarious hits to my API
- Implemented Cross Origin Resource Sharing (CORS) to limit API call to the front end app
- Implemented API KEY to limit access and calls to the API
2025-03-16: Version 2.0.0
- Redesigned the back end of the program to be agentic using LangGraph. The agent will now determine the best source to answer the users question based on what it determines to be the user topic.
- LLM in the Workflow:
- All questions are rewritten to help with search.
- Used to determine the topic and directs towards the appropriate search
- Answers are generated with information from the search passed to the LLM
- Added the BHS event calendar as a data source to answer questions about future events. To limit the amount of information to deal with only events from TODAY until June 30 are considered. June 30 being the last day of school events for the school year 2024-2025.
- Moved profanity filter to frontend to avoid API call
- Update prompts for better responses. Particularly to get better formatted responses with the new calendar data source
- Update front end About section to notify of new source data
- Update the dark mode avatar for better visibility
- Updated the format and wording of the “Hello” message to account for multiple sources
- NOTE: App still can not answer AND questions. Now with multiple sources depending on the question it might pick to search the calendar when it should search the handbook and vice versa. Will address in future releases.
- NOTE: Removed passing original question with rewritten question. Will observe results, but I may need to revert to adding it back to the answer generation. Challenge was managing the original question between states / nodes / edges.
2025-03-09: Version 1.1.0
- Implemented LangSmith to help with debugging and tracing for the application performance
- Updated streamlit to alter icons between dark mode and light mode
- Updated language for disclaimer
- Updated several changes to improve the results of the chatbot
- Updated the chunking strategy based on the formatting of the pages leveraging the that most of the headings in the document are denoted in full caps.
- Updated the embedding model to TogetherAI: BGE (bge-large-en-v1.5)
- Updated vector retrieval search to leverage “fetch_k” to evaluate more chunks prior to returning final “k” documents
- User inputs are now rephrased in context to the handbook to help with semantic search and poorly written questions. Original question is retained to help with and maintain user context
- Rephrased inputs now also help safeguard against abuse of the system
2025-02-28: Version 1.0.0
- Description: The initial iteration of the application is a straight forward Retrieval Augmented Generation (RAG) chatbot. The chatbot uses the Bartlesville High School student handbook as its source of knowledge. The school handbook does contain a few images that this chatbot can not currently use for knowledge such as the School Academic Calendar, Updated Asthma School Health Plan letter and Asbestos Notice Page letter.
- Solution Stack - Backend: FastAPI, Frontend: Streamlit, Host: Railway.com, Vector Database: Chroma, Langchain Framework, TogetherAI LLM: Llama (Llama-3.3-70B-Instruct-Turbo-Free), TogehterAI Embeddings: BERT (m2-bert-80M-32k-retrieval)