Automated hackathon twitter recap agents powered by CrewAI

HackReporter uses CrewAI to orchestrate multiple AI agents that work together to:

1. Analyze hackathon project videos
2. Generate captions and enhance audio
3. Find participant social media profiles using Exa Deep Search AI
4. Create engaging Twitter/X threads
5. Rank videos by engagement potential
6. Automatically create drafts in Typefully for easy scheduling and posting

• kickoff_hackathon_reporter: Process videos in a directory and generate tweet threads
  • Automatically creates a draft in Typefully after thread generation
  • Provides shareable link for review before posting
• create_tweet_thread: Post generated threads via Typefully API (manual trigger)

pip install -r requirements.txt
# or use uv:
uv pip install -r requirements.txt

Create a .env file with:

# AgentOps for monitoring and observability
AGENTOPS_API_KEY=your_agentops_api_key  # Get from https://www.agentops.ai/

# LLM and API Keys
GOOGLE_API_KEY=your_google_api_key
EXA_API_KEY=your_exa_api_key  # Get from https://exa.ai/
CAPTIONS_AI_API_KEY=your_captions_key

# Typefully API for creating tweet drafts
TYPEFULLY_API_KEY=your_typefully_key  # Get from Typefully settings

1. Log in to your Typefully account
2. Go to Settings → Integrations
3. Switch to the Twitter/X account you want to use
4. Generate or copy your API key for that account
5. Add it to your .env file

python server.py

# Process videos
kickoff_hackathon_reporter(
    directory="/path/to/video/directory",
    attendee_list="/path/to/attendees.csv"  # optional
)

# Create tweet thread
create_tweet_thread(tweet_file="output/tweet_thread.md")

HackReporter is integrated with AgentOps for comprehensive monitoring and observability of your AI agents.

• Trace Visualization: See how your CrewAI agents work together in real-time
• Performance Metrics: Track processing time, token usage, and costs
• Error Monitoring: Identify and debug issues in your agent workflows
• Session Recording: Replay entire video processing sessions
• LLM Call Tracking: Monitor all LLM interactions automatically

1. Sign up at https://www.agentops.ai/
2. Get your API key from the settings page
3. Add it to your .env file as AGENTOPS_API_KEY
4. Run the HackReporter - each video processing session will create a new trace
5. View your traces in the AgentOps dashboard

Each trace will be named after the directory being processed and tagged with relevant metadata for easy filtering.

HackReporter automatically creates drafts in Typefully after processing videos:

1. Video Processing: The crew analyzes all videos in the specified directory
2. Thread Generation: Creates a complete Twitter/X thread with:
   • Engaging 3-section title
   • Individual tweets for each project
   • Proper mentions and hashtags
   • Thread numbering (1/, 2/, etc.)
   • Important: Each tweet is separated by 4 blank lines for proper thread splitting
3. Typefully Draft: Automatically creates a draft using the Typefully API
   • Auto-splits the thread at the 4-newline boundaries
   • Generates a shareable link for review
   • Ready to schedule or publish directly from Typefully

• Review Before Posting: Check the thread formatting in Typefully's preview
• Easy Scheduling: Use Typefully's scheduling features
• Media Attachments: Add videos/images to specific tweets in Typefully
• Analytics: Track performance through Typefully's analytics

After processing, you'll receive:

• Tweet thread saved to output/tweet_thread.md (with proper 4-newline formatting)
• Example formatting available in output/example_formatted_thread.md
• Typefully draft ID and shareable link in the console
• Confirmation of successful draft creation

The HackReporter now uses Exa's Deep Search API for intelligent Twitter/X handle discovery:

• Multiple Name Search: Can search for entire teams at once
• Company & Tech Handles: Automatically finds official accounts for mentioned technologies
• Context-Aware: Works even without specific names by analyzing project context
• Graceful Fallbacks: Handles cases where profiles can't be found

# Run the test script to see examples
python test_twitter_search.py

The tool intelligently handles:

1. Missing Names: Uses project context to suggest relevant handles
2. Multiple Team Members: Searches for all team members in one go
3. Tech Stack: Finds official handles for frameworks, tools, and companies mentioned

Agent:

• Role: Video Summarizer
• Goal: Take the content of the video and summarize in in the following tweet style. Here are the examples:
• Tools:
  • Gemini Video API: https://ai.google.dev/gemini-api/docs/video-understanding
  • Features:
    • Automatic handling of video files (inline for <20MB, File API for larger)
    • Audio transcription with timestamps
    • Video clipping (specify start/end times)
    • Custom frame rate sampling (default: 1 FPS)
    • Visual description at key moments
    • Support for mp4, mov, avi, webm, and more formats

Agent:

• Role: Video Captioner
• Goal: Add captions to the video and spellchecks them
• Tools:
  • captions.ai: https://www.captions.ai/add-subtitles

Agent:

• Role: Video Background Noise Remover
• Goal: Remove background noise from the video
• Tools:
  • captions.ai: https://www.captions.ai/add-subtitles

Agent:

• Role: Video Ranker
• Goal: Rank the videos by how interesting they would be to an audience

Agent:

• Role: Person Finder
• Goal: Given the attendee list, do a deep research to find Twitter/X handles for people, companies, and technologies
• Tools:
  • Exa Deep Search API: Advanced AI-powered search for finding social media profiles
  • Features:
    • Handles multiple team members in a single search
    • Finds company and technology official accounts
    • Works even when names aren't available (uses context)
    • Gracefully handles cases where profiles can't be found
    • Assumes tech industry context for better results