AutoResearcher 🤖

Automated Data Collection with LLM, Google Custom Search, Video Transcription, Web Scraping, Instagram API, OpenAI Vision, and OpenAI Whisper.

This repo has kind of turned into a playground for me to use OpenAI LLMs.

This Python project automates data collection on user provided topics.

Features

1. Query Generation: LLM creates relevant search queries.
2. Data Collection: NewsApi, YouTube, and Google Search to collect data on specific topics.
3. Data Scraping: Uses Playwright and BeautifulSoup to fetch and extract web content. Uses MoviePy to extract audio data from videos. Uses OpenAI Whisper Model to create audio transcriptions.
4. Summarization: LLM summarizes scraped data into concise reports.
5. End-to-End Automation: Fully automated from input to summarized output.
6. Instagram Integration:
   • Read a user's Instagram.
   • Process pictures using OpenAI vision.
   • Create an object-weighted graph based on object occurrences in the pictures.
   • Summarize picture settings and subjects.
   • Summarize the user's Instagram in total.
7. Chat Functionality:
   • Chat with LLM
   • Build character from instagram images
   • Chat with instagram character
8. Chat Assistant Timeline Builder:
   • Timeline builder assistant
   • Saves, updates, suggests, finds milestones or memories for a persons life.
   • Uses GPT functions to modify, add, update, delete, find user milestones from milestone store.
   • Allows users to upload images to associate with milestones.
   • Provies output for visualizing on a timeline in the front end.

Technologies

• Python, OpenAI API, Google Custom Search API
• Playwright, BeautifulSoup, MoviePy, NewsAPI
• InstagramAPI, NetworkX

Setup

1. Clone the repo:

   git clone https://github.com/your-username/your-repo-name.git
   cd your-repo-name

2. Install dependencies:

   pip install -r requirements.txt

3. Set up API keys:
   Create a .env file with:

   OPENAI_API_KEY=your_openai_api_key
   GOOGLE_CUSTOM_SEARCH_API_KEY=your_google_custom_search_api_key
   GOOGLE_CX=your_google_cse_id
   NEWS_API_KEY=newsapi_key
   INSTAGRAM_CLIENT_ID=insagram_client_id
   INSTAGRAM_CLIENT_SECRET=insagram_client_secret
   INSTAGRAM_REDIRECT_URI=instagram_redirect_uri
   INSTAGRAM_ACCESS_CODE=insagram_access_code

4. Run the script: Currently only the "news" source is supported.

   python src/researcher/main.py --topic "Your topic here" --source "[news, google, youtube, all]"

5. Check Results Folder: Find result json file in results folder

Usage

1. Input a topic.
2. LLM generates search queries.
3. Data source retrieves results.
4. MoviePY transcribes video or Playwright & BeautifulSoup scrapes web pages.
5. LLM summarizes the scraped content.

Future Enhancements

• Support for more search engines.
• Advanced query generation and filtering.
• Customizable summarization options.
• Add UI for visualizing results
• Add subscription features for users to get updates on a schedule

Needed Improvements

• Web page scraping in web_page_reader.py
• More abstract data_gatherer.py class
• Chunking in openAI requests for large text especially in video summarization
• Potentially increase performance by decreasing run time?
• Add SerpApi
• Reduce number of LLM calls and increase relevancy of sumary data