Inspiration
Insight Mentor was inspired by a very real problem I—and many other college students—face every semester: studying takes forever.
Not because the material is hard, but because processing it takes hours.
Dense PDFs. Endless slides. Chapters full of fluff.
It often felt like I spent more time organizing information than actually learning it.
I wanted something that would:
- Clean the clutter
- Extract what matters
- Teach concepts clearly
- And guide me with a personalized study plan
What it does
This idea turned into Insight Mentor—an AI-powered tool that transforms raw study materials into clean notes, flashcards, quizzes, and interactive mentorship.
How we built it
Insight Mentor is built with:
- Python for backend logic
- Streamlit for a fast, interactive UI
- OpenAI + Gemini for flexible model selection
- PyPDF / Unstructured for file parsing
- Local vector store for embeddings and retrieval
- Markdown + GitBook for documentation
Below is an overview of the project's architectural structure:
│ User (Student) │
│ - Uploads files │
│ - Clicks Generate │
│ - Uses Mentor / Quizzes │
└─────────────┬───────────────┘
│
▼
┌────────────────────────────────┐
│ Streamlit UI │
│ - Upload area │
│ - Notes / Flashcards view │
│ - Study plan / Mastery view │
│ - Prompt Coach + Mentor chat │
└─────────────┬──────────────────┘
│
┌────────────────────┼────────────────────────┐
▼ ▼ ▼
┌─────────────────┐ ┌───────────────────┐ ┌──────────────────┐
│ File Ingestion │ │ Study Assets │ │ Mentor Mode │
│ & Extraction │ │ Generator │ │ (Q&A Engine) │
└───────┬──────────┘ └─────────┬────────┘ └─────────┬────────┘
│ │ │
▼ ▼ ▼
┌─────────────────────┐ ┌─────────────────────┐ ┌─────────────────────┐
│ Text Cleaning & │ │ Notes Generator │ │ Prompt + Contextual │
│ Semantic Chunking │ │ Flashcard Generator │ │ Retrieval │
│ - remove headers │ │ Quiz Generator │ │ - selects relevant │
│ - normalize lists │ └─────────────────────┘ │ chunks only │
│ - chunk paragraphs │ └─────────┬───────────┘
└──────────┬──────────┘ │
│ │
▼ ▼
┌─────────────────────┐ ┌─────────────────────────┐
│ Local Vector Store │◀────────────────────────▶│ LLM Provider Layer │
│ - embeddings index │ │ (Model Switcher) │
│ - per-document │ │ │
└──────────┬──────────┘ │ 🔹 OpenAI (deep, rich) │
│ │ 🔹 Gemini (fast, cheap)│
▼ │ - automatic fallback │
┌─────────────────────┐ └───────────┬────────────┘
│ Concept Tagging & │ │
│ Mastery Model │◀──────────────────────────────────────┘
│ - detects concepts │
│ - stores mastery % │
│ - powers study plan │
└──────────┬──────────┘
│
▼
┌─────────────────────┐
│ Study Plan Builder │
│ - ranks weak topics │
│ - creates tasks │
│ - estimates time │
└─────────────────────┘
Challenges I ran into
During the 48 hours, these were the problems I ran into with the project:
PDF Extraction Issues
- Some PDFs had weird formatting, missing text, or repeated headers
- Solved with aggressive cleaning + better chunk detection
Chunking/ Structure Detection
- Ensuring the AI only learned from relevant text
- Preventing hallucinations by grounding prompts in extracted chunks
Multi-Model Management
- Handling different API styles (OpenAI vs Gemini)
- Creating a smooth fallback system when one model failed
UI/ UX Design
- Making sure the interface didn’t overwhelm new users
- Creating clear sections for notes, flashcards, mentor chat, quizzes, and study plans
⏱️ Time Crunch
This was built in less than 48 hours, which made scoping, prioritizing, and execution very intense.
Accomplishments that we're proud of
Honestly, the biggest accomplishment was simply just building something meaningful—a tool that I know students will genuinely find beneficial.
Despite the time pressure and technical challenges, I’m proud that Insight Mentor became:
- Something that reduces stress and saves hours for students
- A tool that reflects the struggles I personally faced while studying
What I learned
AI & Prompt Engineering
- How to design prompts that generate structured notes, hierarchical summaries, and high-quality flashcards
- Understanding the trade-offs between Gemini and OpenAI for reasoning vs. speed
Product Design
- How to design a simple, clean UI that feels approachable for students
- The importance of limiting cognitive load with careful layout and progressive steps
Backend & Architecture
- File extraction, chunking, and concept tagging
- Storing embeddings locally and retrieving relevant chunks
- Managing multiple LLM providers with fallback and error-handling
Math / Models
I also explored mastery scoring, starting each concept with a baseline of
[
\text{Mastery}_{0} = 0.30
]
and incrementing based on quizzes, flashcards, and correct reasoning—similar to spaced repetition systems.
What's next for Insight Mentor
What’s next? I’m still exploring the possibilities, but I’m excited to keep researching and building Insight Mentor into something students will genuinely love and truly rely on.
Log in or sign up for Devpost to join the conversation.