# Project export: Tailwind
This document was generated by HackStack to give an AI agent context about a hackathon project. Sections are labeled with their provenance; content marked as truncated was cut to keep this document small.
## Project metadata
- Hackathon: CruzHacks 2025
- Tagline: Have you ever felt like your midterm or final review wasn't enough, tailwind has got you covered
- Devpost: https://devpost.com/software/tailwind
- GitHub: https://github.com/waterball01/cruzhacks25
- Video: https://www.youtube.com/embed/Hy-u4N3ZAVw?enablejsapi=1&hl=en_US&rel=0&start=&version=3&wmode=transparent
- Team: 2 GitHub contributor(s) — waterball01 (10 commits), ServeshKarnawat (4 commits)
## Devpost submission (written by the team)
### Inspiration
Have you ever spent hours if not days studying material for an exam only for the exam to be nothing like the study material. We have countless times and we wanted to come up with a solution to provide the perfect study material and extra practice directly based on the professors lectures.
### What it does
Tailwind creates study material directly based on professor's materials using RAG and AI to prepare students for exams. Students can submit their professors lecture notes, number of questions, and specific topics they want covered, and Tailwind creates practice problems from this information and then analyzes the answers to see how correct they are and gives feedback on the user's answers.
### How we built it
We made this using a flask, html and css website which calls gemini to extract text from pdfs or images and then breaks that text into chunks which we use chromadb (a RAG framework) to query and get the most related information to a specific question, we then call gemini again to generate questions regarding the topics, and then collect answers from the user. We call gemini one last time to provide feedback on the responses and answer any followup questions the user may have.
### Challenges we ran into
We had issues with data formatting, indexing, flask, html, and debugging. Neither of us were very familiar with html integration with Flask so we had issues trying to figure out how things translated between the languages of html and python. We also had limited experience with the gemini api so understanding how it worked and making it do what we wanted was a little difficult. When using the gemini api, it took a long time to generate questions and answers so debugging it was difficult as we had to wait a long time when trying to figure out what we did wrong.
### Accomplishments we're proud of
Finishing with a fully functional AI powered tool that can help with studying for all people. Becoming much better at HTML and flask. Gaining experience integrating AI into web development
### What we learned
We got more familiar with Web Development, using AI like Gemini, and database management.
### What's next
We are going to add a special user interface for True/False and multiple choice questions. We are also going to improve the AI model of the chatbot used to answer further questions AI answer feedback. We also want to optimize our use of gemini so we get faster responses as our current ones are a little slow.
## README (from the GitHub repository)
# Tailwind
An AI-powered study assistant built at CruzHacks 2025. Upload your lecture notes
or handwritten notes and Tailwind will automatically generate quizzes to help you
study smarter.
## Features
- Upload typed or handwritten lecture notes
- AI-generated quiz questions based on your content
- Clean web interface for a smooth study experience
- Fast turnaround — upload and start quizzing in seconds
## Tech Stack
- **Frontend:** HTML, CSS
- **Backend:** Python
- **AI:** LLM-powered question generation from uploaded content
## Getting Started
### Prerequisites
- Python 3.8+
- pip
## Detected evidence (automated analysis)
Indexed codebase: 8 recognized source files, 24 KB.
- CSS (language) — detected in the code
- HTML (language) — detected in the code
- Python (language) — detected in the code
- Flask (technology) — claimed on Devpost, not found in the code
- Google Gemini (technology) — claimed on Devpost, not found in the code
## Codebase structure (from repository index)
### Files (8 of 8)
```
Cruzhacks2025/ai_core.py
Cruzhacks2025/app.py
Cruzhacks2025/config.py
Cruzhacks2025/static/tailwind.css
Cruzhacks2025/templates/index.html
Cruzhacks2025/templates/quiz.html
Cruzhacks2025/templates/results.html
README.md
```
### Dependencies
No dependency index available.
### Recent commits (newest first)
- Update README.md
- finsihed????
- multiple files
- fixed prompt again
- fixed prompt
- merged ai with website
- Delete Cruzhacks2025/testGemini
- Add files via upload
- Delete Cruzhacks2025 directory
- ai_core parts completed
- merged files together
- new chroma
- Add files via upload
- Initial ai stuff
- Initial commit
## Key source files (fetched from GitHub, selected and truncated for size)
### Cruzhacks2025/app.py
```python
from flask import Flask, render_template, request, redirect, url_for, session
from werkzeug.utils import secure_filename
import os
from ai_core import generate_study_questions, evaluate_answers, answer_question
from flask import jsonify
import os
from sentence_transformers import CrossEncoder
from chromadb.utils.embedding_functions import SentenceTransformerEmbeddingFunction
import google.generativeai as genai
from config import genai, chroma_collection, cross_encoder, Base
app = Flask(__name__)
app.secret_key = "supersecretkey"
app.config['UPLOAD_FOLDER'] = 'uploads'
os.makedirs(app.config['UPLOAD_FOLDER'], exist_ok=True)
@app.route("/", methods=["GET", "POST"])
def index():
if request.method == "POST":
files = request.files.getlist("file")
file_paths = []
for file in files:
if file and file.filename:
filename = secure_filename(file.filename)
file_path = os.path.join(app.config['UPLOAD_FOLDER'], filename)
file.save(file_path)
file_paths.append(file_path)
question_count = int(request.form["question_count"])
prompt = request.form["prompt"]
session["file_paths"] = file_paths
session["question_count"] = question_count
session["prompt"] = prompt
print("FILES:", file_paths)
questions = generate_study_questions(file_paths, prompt, question_count, session_id=90210)
session["questions"] = questions
return redirect(url_for("quiz"))
return render_template("index.html")
@app.route("/quiz", methods=["GET", "POST"])
def quiz():
questions = session.get("questions", [])
if request.method == "POST":
answers = [request.form.get(f"answer_{i}") for i in range(len(questions))]
session["answers"] = answers
return redirect(url_for("results"))
return render_template("quiz.html", questions=questions)
@app.route("/results")
def results():
questions = session.get("questions", [])
answers = session.get("answers", [])
feedback = evaluate_answers(questions, answers,session_id=90210)
qa_pairs = zip(questions, answers, feedback)
return render_template("results.html", qa_pairs=qa_pairs)
@app.route("/clarify", methods=["POST"])
def clarify():
data = request.get_json()
question = data.get("question", "")
answer = answer_question(question, session_id=90210)
return jsonify({"answer": answer})
if __name__ == "__main__":
app.run(debug=True)
```
### Cruzhacks2025/config.py
```python
import os
import chromadb
from chromadb.utils.embedding_functions import SentenceTransformerEmbeddingFunction
import google.generativeai as genai
from sentence_transformers import CrossEncoder
from sqlalchemy.orm import declarative_base
# --- Gemini API setup ---
genai.configure(api_key=os.environ["GEMINI_API_KEY"])
# --- ChromaDB setup ---
embedding_function = SentenceTransformerEmbeddingFunction()
chroma_client = chromadb.Client()
chroma_collection = chroma_client.get_or_create_collection(
name="output",
embedding_function=embedding_function
)
# --- Cross-encoder setup ---
cross_encoder = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2')
# --- SQLAlchemy base setup ---
Base = declarative_base()
# --- Export for import in other modules ---
__all__ = [
"genai",
"chroma_collection",
"cross_encoder",
"Base"
]
```
### Cruzhacks2025/ai_core.py
```python
from PIL import Image
from chromadb.utils.embedding_functions import SentenceTransformerEmbeddingFunction
import google.generativeai as genai
import uuid
from sqlalchemy import create_engine, Column, String, Text
from sqlalchemy.orm import sessionmaker
from datetime import datetime, timezone
from pdf2image import convert_from_path
from config import genai, chroma_collection, cross_encoder, Base
class ChatHistory(Base):
__tablename__ = 'chat_history'
id = Column(String(36), primary_key=True)
session_id = Column(String(36))
role = Column(String(10))
message = Column(Text)
created_at = Column(String(20), default=lambda: datetime.now(timezone.utc).isoformat())
def init_db():
engine = create_engine("sqlite:///chat.db", future=True)
Base.metadata.create_all(engine)
Session = sessionmaker(bind=engine, future=True)
return Session()
def save_message(db_session, session_id, role, message):
chat_entry = ChatHistory(
id=str(uuid.uuid4()),
session_id=session_id,
role=role,
message=message
)
db_session.add(chat_entry)
db_session.commit()
def load_chat_history(db_session, session_id):
history = db_session.query(ChatHistory).filter(
ChatHistory.session_id == session_id
).order_by(ChatHistory.created_at).all()
return [{"role": entry.role, "parts": [entry.message]} for entry in history]
def extract_from_pdf(filepath):
pages = convert_from_path(filepath)
words = []
for page in pages:
model = genai.GenerativeModel('gemini-2.0-flash')
response = model.generate_content(["Extract all the text from this image, do not include any other words oustide the text:", page])
words.append(response.text)
return "\n".join(words)
def extract_text_from_file(filepath):
print(filepath + "\n\n\n\n")
print(str(filepath.rstrip().lower().endswith('.pdf')) + "\n\n\n\n")
if filepath.rstrip().endswith('.txt'):
with open(filepath, 'r') as f:
return f.read()
elif filepath.rstrip().lower().endswith('.pdf'):
print("\n\n\nAAAAAAAAAAAAAAAA\n\n\n\n")
return extract_from_pdf(filepath)
elif filepath.rstrip().lower().endswith(('.png', '.jpg', '.jpeg')):
image = Image.open(filepath)
model = genai.GenerativeModel('gemini-2.0-flash')
response = model.generate_content(["Extract all the text from this image, do not include any other words oustide the text:", image])
return response.text
else:
raise ValueError("Unsupported file format")
def chunk_text(text, max_len=500):
paragraphs = text.split("\n")
chunks, current = [], ""
for para in paragraphs:
if len(current) + len(para) < max_len:
current += para + " "
else:
chunks.append(current.strip())
current = para + " "
if current:
chunks.append(current.strip())
return chunks
def gemini_chat(session_id, question, db_session):
history = load_chat_history(db_session, session_id)
model = genai.GenerativeModel('gemini-2.0-flash-lite', system_instruction="You are a helpful expert tutor. Your users are asking questions about information provided by their lecture notes or other sources regarding their class. You will be shown the user's question, and the relevant information, answer the user's question using only this information.")
chat = model.start_chat(history=history)
response = model.generate_content(question,generation_config=genai.types.GenerationConfig(
candidate_count=1,
temperature=0.3,
),)
save_message(db_session, session_id, "user", question)
save_message(db_session, session_id, "model", response.text)
return response.text
def chroma(filepath, chroma_collection):
text = extract_text_from_file(filepath)
chunks = chunk_text(text)
ids = [str(uuid.uuid4()) for _ in chunks]
chroma_collection.add(ids=ids, documents=chunks)
return chunks
def retrieve(query, chroma_collection, n_results=10):
result = chroma_collection.query(query_texts=[query], n_results=n_results, include=["documents"])
docs = result["documents"][0]
scores = cross_encoder.predict([[query, doc] for doc in docs])
reranked = [doc for _, doc in sorted(zip(scores, docs), key=lambda x: x[0], reverse=True)]
return reranked
def generate_study_questions(filepaths, specification, numqs, session_id):
query = "Generate study questions based on the following lecture notes. Only include information from the lecture notes, do not include any other information. Make sure to hit on all major points. Make sure to have" + str(numqs) + " questions. The user specfically requested the following: " + str(specification) + ", if it does not make sense ignore it. Do not include any oother characters aside from the question itself, do not include the question numbers. Add a new lince charcter between each question.\n\n"
db_session = init_db()
for filepath in filepaths:
chroma(filepath,chroma_collection)
retrieved = retrieve(query, chroma_collection)
combined_info = "\n".join(retrieved)
response = gemini_chat(session_id, f"{query}\n\nContext:\n{combined_info}", db_session)
data = response.split('\n\n')
db_session.close()
return data
def evaluate_answers(questions, answers, session_id):
query_prefix = (
"Correct the following questions and answers. For each one, include:\n"
"- Feedback on the given answer\n"
"- The correct answer\n"
"- A brief explanation of how to get the correct answer\n"
"- Do not include the question itself simply follow the following format for each question\n"
"Feedback: \n"
"Correct Answer: \n"
"Explanation: \n"
"Include a newline in between responses to different que
[truncated — 2011 more characters]
```
### Cruzhacks2025/templates/quiz.html
```html
Quiz
