Nova Lecture AI

NovaLecture AI: The Story Behind the Project

Inspiration

The genesis of NovaLecture AI came from a simple observation: students and professionals are drowning in video content. With the explosion of online learning, long-form lectures have become the standard, yet our ability to process and retain that information hasn't kept pace. We wanted to build a tool that doesn't just "summarize," but actively transforms passive viewing into active learning.

Our goal was to create a bridge between the Amazon Nova 2 Lite model’s reasoning capabilities and the practical needs of a student: summaries for quickly catching up, flashcards for active recall, and visual aids for deeper understanding.

How We Built It

The architecture of NovaLecture AI is designed for speed and modularity:

1. The Brain: Amazon Bedrock & Nova Everything centers around Amazon Nova 2 Lite. We utilized the Bedrock Converse API to perform several distinct tasks:

Summarization: Transforming transcripts into structured notes. Structured Extraction: Using JSON mode to generate 5 MCQs, 5 flashcards, and 3 short answers for every lecture. Creative Scripting: Designing slide-by-slide layouts for PowerPoint and scene-by-scene scripts for animated videos.

1. The Engine: Python & FastAPI The backend handles the "heavy lifting":

Transcription: We integrated faster-whisper for near-real-time audio-to-text conversion. Media Generation: To make study kits truly "visual," we used python-pptx to programmatically build presentations and moviepy combined with Google Text-to-Speech (gTTS) to render animated explainer videos.

1. The Interface: React & Glassmorphism We built a modern, responsive frontend using React. We focused on a "premium" feel inspired by modern SaaS designs—using glassmorphism, subtle animations, and a strict dark-mode palette to keep the focus on learning.

AWS Services Used

NovaLecture AI is powered by the Amazon Web Services (AWS) ecosystem:

Amazon Bedrock: The primary orchestration service for securely accessing foundation models. Amazon Nova 2 Lite: The "brain" used for lightning-fast transcript summarization and study material generation. AWS IAM: Handles secure authentication between the local backend and Bedrock. AWS SDK for Python (Boto3): Used for seamless integration of Bedrock's API into our Python engine.

What We Learned

We delved deep into the mechanics of LLM prompt engineering. One of the most interesting aspects was balancing token usage. For instance, we learned that for long transcripts, we could model the information density as: $$ I = \sum_{i=1}^{n} \frac{T_i}{V} $$ Where $I$ is the information density, $T_i$ are the key tokens, and $V$ is the video length. By optimizing the context window for Nova 2 Lite, we ensured high-quality summaries even for hour-long lectures.

Challenges We Faced

Building this in a hackathon environment brought unique hurdles:

Bot Detection: Initially, downloading YouTube videos for processing was blocked by automated bot detection. We solved this by implementing cookie-based authentication within our yt-dlp wrapper. Processing Latency: Analyzing an hour-long video can take time. To improve user experience, we implemented granular status updates (e.g., "Transcribing...", "Summarizing...") and background pre-generation, so study materials are ready the moment the summary is finished. Local vs. Cloud Environments: We encountered an issue where Microsoft’s Office Viewer (used for PPTX previews) cannot access localhost. We adapted by adding an automated environment detection system that offers direct downloads when running locally, ensuring the UI never feels "broken."

The Future of NovaLecture

NovaLecture AI is just the beginning. We envision a world where every lecture is automatically transformed into a personalized virtual tutor, making education more accessible, engaging, and efficient for everyone, everywhere.

Built With

• amazon-web-services
• bedrock
• novaai
• python
• react