Mötley Crëw PRESENTs
Real-time AI-powered podcast assistant that auto-generates visuals, fact-checks, and provides contextual insights on an infinite TLDraw canvas.
Project Description
PRESENT: Intelligent Conversation Augmentation for Live Podcasts
PRESENT is an AI-driven production assistant that supercharges live conversations—think “Jamie” from Joe Rogan, but powered by multi-agent AI. Designed for podcasters and streamers, PRESENT listens to live audio, surfaces verified information, and injects relevant media (like video clips, profiles, and fact-cards) directly into your interactive show UI—without interrupting the flow.
🧠 How It Works
• Live Audio: Listens via a LiveKit-backed audio stream, transcribing in real time using OpenAI Whisper.
• Multi-Agent Crew: Uses CrewAI (agents defined in YAML) to orchestrate a team of specialist agents:
• Fact Checker (validates claims using web search tools)
• Profile Researcher (deep-dive dossiers using Perplexity API)
• YouTube Analyzer (finds relevant videos, extracts transcripts/timestamps)
• Context Provider (fetches background and definitions)
• Conversational Intelligence: A central Conversational Assistant (GPT-4-powered) decides when to intervene, keeping the AI silent unless its input genuinely adds value.
• Dynamic Presentation: Results are rendered on an interactive front-end (Next.js + Tambo), which supports both chat threads and a drag-and-drop infinite canvas for pinning facts, clips, and graphs during the show.
• API-First: All outputs stream over SSE/REST APIs, enabling integration with overlays, Discord bots, or external dashboards.
🚀 Key Features
• True Real-Time: Audio-to-transcript-to-agent pipeline operates live, <1s latency.
• Rich UI: Embeds YouTube clips, displays citations, and lets hosts rearrange fact-cards on a collaborative canvas.
• Deep Research: The flagship Profile Researcher agent rapidly assembles comprehensive dossiers on people, topics, or organizations—context that used to take hours of Googling.
• Extensible & Modular: Add new agents or data sources in minutes by editing YAML, with all infra ready for more specialist roles.
• Fact-Checking on the Fly: Fact-checker and context agents validate claims and surface sources without interrupting the speaker.
🛠 Tech Stack
Layer Tools & Libraries Role
AI Agents CrewAI (YAML-defined) Orchestrate and coordinate agents
Audio LiveKit, livekit-agents, Whisper, Silero Streaming, VAD, speech-to-text
LLM Reasoning OpenAI GPT-4 Plan generation, context understanding
Research Perplexity API, SerperDev, YouTube API Deep search, web scraping, video analysis
Front-end Next.js, Tambo, Tailwind CSS Chat feed, infinite canvas, live updates
Back-end FastAPI (Python), SSE/REST API Streams agent results, task control
Developer Experience Codespaces, dotenv, ESLint, Prettier Modern, collaborative workflow
🌟 What Makes PRESENT Unique
• From Conversation to Knowledge Canvas: Go beyond a chat window—hosts can arrange facts, media, and summaries spatially, turning static talk into a visual knowledge base.
• No Interruptions, Just Augmentation: The AI only “speaks” when it adds value—never interrupting, always assisting.
• Easy Integrations: Plug PRESENT’s agent outputs into any external system via APIs.
• Roadmap: Future releases will expand agent skills (e.g. auto-generated video highlights, smarter “interruption” timing, even better semantic search).
📈 Real-World Impact
PRESENT eliminates the need for on-air research breaks. It instantly arms hosts with facts, media, and context—keeping the conversation flowing and the audience better informed. It’s a tireless digital producer that works behind the scenes, turning every podcast or broadcast into a knowledge-rich experience.
⸻
Source & Demo:
GitHub Repository — Ready to clone and run.
Get started: docker compose up and see PRESENT in action.
Prior Work
Scrapped my initial tldraw MVP