All projects

ClarkuBot

One chat box for everything scattered across Clark University’s websites, built in a 24 hour hackathon.

Client
IEEE Clark Hack 2025
Duration
24 hours
Stack
Python, Gemini 2.0 Flash, google‑genai, Streamlit
ClarkuBot architecture: Streamlit app.py calls main.py, which initialises the Gemini client and uploads CSV and TXT data sourcesClarkuBot architecture: Streamlit app.py calls main.py, which initialises the Gemini client and uploads CSV and TXT data sources
System architecture.

Course listings, career services and campus pages at Clark live on many separate sites. ClarkuBot scrapes them into one knowledge base and lets students just ask.

A Streamlit front end talks to a Python back end that grounds Google’s Gemini 2.0 Flash in about 3.4 MB of scraped university text and course spreadsheets.

How it works

  1. Load the knowledge

    On start, every CSV and TXT file in the project folder is uploaded to Gemini and attached to the conversation, so answers come from Clark’s own pages.

  2. Answer with guardrails

    Temperature 0.5 keeps answers factual, and Gemini’s safety filters block harmful content at the strictest level.

  3. Remember the conversation

    Streamlit session state keeps the chat history and the uploaded files, so follow up questions keep their context and files upload only once.

Sequence diagram of a ClarkuBot session from first load to streamed Gemini responseSequence diagram of a ClarkuBot session from first load to streamed Gemini response
How a request flows through the system.

View the code on GitHub