Hack America 2026

The Submission

I submitted Study Assistant to the 2026 Hack America Hackathon. A project I built for myself first and foremost — I put a lot of time into it, from easy installation to performance on low-end hardware, and expanded it to support cloud models (Groq, ChatGPT, Anthropic) alongside local Ollama inference.

View on GitHub View Project Page

Demo

Setup

A setup.sh script handles everything on Linux, macOS, and Windows. First grant execute permission, then run it:

Windows (PowerShell):

chmod +x setup.sh
bash setup.sh

macOS / Linux:

chmod +x setup.sh
./setup.sh

The script creates a .venv virtual environment automatically. Activate it, then run the program:

Windows:

.venv/bin/Activate.ps1
python run.py

macOS / Linux:

source .venv/bin/activate
python run.py

No GPU? No problem — skip Ollama and use an API key from Groq, Anthropic, or another provider. Run the CLI or Web UI settings and select the proper key to save.

About the Project

Inspiration: Built out of frustration with tools like Quizlet and NotebookLM. I have ADHD and strongly detest studying, so I wanted a tool that could ingest my own notes and help me learn material to a depth those tools don't reach — and do it faster.

What it does: Accepts Markdown files, PDFs, and images as input and uses a local RAG pipeline to extract and index the content. You can then query your notes conversationally — the assistant retrieves the most relevant chunks from ChromaDB and responds using a local Ollama model (or a cloud model of your choice). Available as both a CLI and a Streamlit Web UI. The Web UI additionally supports direct camera uploads for photographing physical notes.

How it was built: Python is the core language. ChromaDB handles the vector database layer for semantic retrieval. Ollama runs local LLM inference so nothing leaves your machine. Streamlit powers the Web UI and unlocks the camera upload feature.

Challenges: Staying motivated on a project longer than a week is always a challenge. The temptation to let AI write everything was real — I used it only for debugging and learning how ChromaDB works under the hood.

What I learned: ChromaDB's vector storage and query model, and how to run local LLM inference via Ollama — both of which I plan to bring into the Finance Kit.

Status: Actively in development. Check the GitHub repo for updates.

Features

Chat

Query your notes in natural language. The LLM answers from your ChromaDB content and saves artifacts like flashcards and quizzes for later review.

Add Content

Upload files or take a photo directly in the Streamlit UI to add to ChromaDB. Without this, the knowledge base never grows.

Setup Script

One shell script initialises the entire project — virtual environment, dependencies, everything. No manual pip installs or config fiddling.

Config.toml

Stores API keys for cloud AI engines securely. Switching providers is a single setting change via the CLI or Web UI.

Auto-Update

Checks for updates from the GitHub repo so you never have to manually reinstall. Available in both the CLI and the Web UI.

Cloud Model Support

Groq, ChatGPT, Anthropic, and more — for machines without a GPU or strong CPU. Fully interchangeable with Ollama via the settings interface.

← Back to Hackathons Back to top