Pitch, Please
A submission to UofTHacks 12.
🧚Inspiration🧚
Pitches are a challenging part of every hackathon -- after a whole weekend of sleep-deprived hacking, you're now faced with the task of showcasing all of your hard work in just a few minutes to judges with various technical backgrounds. How can you ensure your ideas come across clearly from each judge's perspective and maybe secure a cool prize?
📈Introducing Pitch, Please📈
Pitch, Please is a live pitch simulator that helps hackers build the skills they need to showcase their work. It provides fresh perspectives from realistic judges with unique personalities to help hackers flush out a compelling pitch. Hackers can upload the rubric they expect to be judged on, list the sponsors who will be judging them, and jump right into a live run-through of the pitch. Following the live pitch, hackers will engage in a Q&A session with the judges and receive feedback based on their performance, just like the real deal.
Hackathons aren't the only place people pitch ideas -- Pitch, Please would be a great tool for startup founders looking to pitch their next big idea. Our platform is uniquely adaptable to these different scenarios:
For Hackathons:
- Upload the provided judging rubric for automatic analysis
- Select specific sponsor companies whose representatives will be judging
- Our system generates judge personas based on deep analysis of the sponsor companies, including their technical focuses, company values, and ideal employee profiles
- Experience realistic pitch simulations with judges who understand their company's perspective
For Startups:
- Our AI agents automatically generate tailored investment rubrics by analyzing:
- Venture capital firms' investment histories
- Partners' ideologies and public statements
- Portfolio company patterns
- Investment thesis data
- Success criteria from previous investments
- Create precise judge personas that reflect specific investors or VC firms
- Practice pitching to investors who understand your target firms' investment strategies
What sets Pitch Please apart is its dynamic judging panel that simulates realistic deliberation. Our AI judges don't just evaluate independently - they engage in detailed discussions, debate points, and work together to reach consensus scores, just like real judging panels. Each judge maintains their unique perspective and company values while collaborating to provide comprehensive feedback.
User flow:
- Upload a picture or screenshot of a rubric you'd like to be judged on for hackathons, or target investors for startups.
- Include a list of sponsors or target judges you expect to pitch for.
- Live pitch! Give it your best go.
- Interact with the judges in a live Q&A session based on your pitch.
- Receive detailed feedback on your performance, including areas of strength and weakness.
- Rinse and repeat!
⚙️How we built it⚙️
Frontend: Next.js/React powering a responsive UI with real-time video streaming, dynamic feedback displays, and seamless WebSocket integration for live interactions
AI Judges: Langchain orchestrating sophisticated judge personas through OpenAI, each with unique personalities and expertise reflected in their questioning and feedback styles
Backend: Python FastAPI handling complex async operations, managing WebSocket connections, and coordinating real-time video/audio processing
OCR: Pytesseract processing uploaded rubric images to automatically extract and structure evaluation criteria
Web Scraping: Custom scripts to gather and analyze company and investor data for generating tailored rubrics and personas
Additional Technologies: DeepFace for emotion analysis during presentations, WebSocket protocol for real-time communication, and various audio processing libraries for speech analysis
⛰️Challenges & Accomplishments⛰️
We had a few technical hurdles to cross, including reducing the AI judges' response latency and seamlessly streaming video and audio to process the data quickly. Although these weren't critical in our demo for this weekend, they are things we'd love to tackle as next steps!
In general, our project involved a relatively complex backend to handle live pitching and generating meaningful feedback from multiple perspectives. It was a challenge to build and string everything together over the course of the hackathon. We're proud of creating a system that successfully integrates these complex elements into a realistic and useful application.
🛣️What’s next for Pitch, Please🛣️
- Reduce latency in AI agent responses.
- Extract more metrics from the pitch for more detailed feedback.
- Implement more features to support different formats of pitches (eg. startups).
- Host it somewhere to see it in action outside of our local environment!
- Add support for team presentations with multiple speakers.
- Expand our investor database and analysis capabilities for startup pitches.
🔗Links🔗
GitHub: https://github.com/whyismynamerudy/pitch-please/tree/main
Video demo: https://www.youtube.com/watch?v=L5SVBExqQyU