carlkho-minerva / 24sumr_edgur-features

Edgur Features Repository

This repository stores various features developed for Edgur, aimed at enhancing user experience and functionality.

1. Recommender Algorithm

The recommender algorithm utilizes Pandas Chroma DB and a uniform probability approach to recommend videos based on user interests. Here's how it works:

• Initialization: The algorithm initializes a queue for user interests, which includes topics ranging from a minimum of four to a maximum of eight.

• Topic Weight Distribution: Topic weights are distributed among each other to ensure each topic has its space in the partition.

• Recommendation Generation: Using a random number generator (random.randint), the algorithm selects a video based on the generated number.

• Interaction Tracking: Upon user interaction with the recommended video, actions such as sharing and looping are observed. If a new tag is encountered, it's added to the user's interest dictionary.

• Queue Management: Videos are cached in advance to ensure they are readily available when users scroll rapidly, preventing loss from the queue.

• Chroma DB Integration: Chroma DB handles the vector storage part, utilizing OpenAI's embedding model (specifically OpenAI Text Embedding 3 Small) to convert words into vectors. This integration ensures that the algorithm returns the most relevant video given a specific tag while skipping videos already watched using a watched_videos dictionary.

2. AI Search

AI Search enhances traditional search functionalities by offering custom playlists or mini video courses based on existing videos in the database. Here's how it functions:

• Query Processing: Using POM engineering, the algorithm extracts key details from user queries to generate a curriculum of potential video titles or beginner test cues.

• Chroma DB Utilization: Each curriculum bullet is processed through Chroma DB to identify the most relevant video based on vector embeddings, ensuring precise content retrieval.

3. AI Quiz Generation

The AI Quiz feature dynamically generates multiple-choice questions from video transcripts, offering an interactive learning experience. Here's a breakdown:

• Question Generation: Utilizing OpenAI's API, the feature creates quiz questions directly from the video content.

• Answer Formatting: The generated questions include four possible answers, with one correct answer and three incorrect answers formatted as '~ Answer 1 ~ Answer 2 ~ Answer 3 ~ Answer 4'.

• Answer Randomization: Answers are shuffled to ensure a randomized order, enhancing user engagement and learning retention.

4. AI Summary on Timeout

This feature provides concise summaries of videos based on user session duration, offering quick insights into watched content. Key features include:

• Automatic Summary Generation: Summaries are pre-generated upon video upload, enabling efficient lookup using hash maps or dictionary operations during user sessions.

5. Whisper on the Web

Whisper on the Web introduces a microphone button for text input, enhancing user interaction with question-and-answer cards and search functionalities. It facilitates user engagement by allowing voice-based input for searches.

6. Revenue Calculation for Stripe Payouts

This feature calculates revenue distributions for Stripe Payouts API via Stripe Connect -> content creators based on Edgur's monthly revenue, inspired by models like Spotify's stream shares. It integrates business operations with revenue tracking and vendor payouts, ensuring transparent and efficient financial management.