Submission by Elia

Pitch Perfect -> Practice your VC pitch with the actual investor, before you ever walk in the room. You type a VC partner's name. Firecrawl runs a parallel multi-source research pipeline - scraping their essays, podcast transcripts, firm website, and portfolio patterns. The output is a rich investor profile: thesis, red flags, communication style, and the questions they're publicly known to ask. Then ElevenLabs becomes that investor. Using a dynamic system prompt override at session start, the agent is re-initialized with the Firecrawl research injected directly into its persona. Every session is a genuinely custom character - not a generic VC. The AI asks questions grounded in the investor's actual public record. Founders do a double-take. Firecrawl usage: 5 parallel sources per session: Medium/Substack essays, podcast transcripts, firm website, personal blog. Structured extraction via `search()` + `scrapeUrl()`. ~847 data points per profile. ElevenLabs usage: Conversational AI with `overrides.agent.prompt.prompt` injected at session start. Dynamic persona for every founder, every pitch. Also used for TTS investor briefing audio. The technical hook: most submissions use the ElevenLabs agent as-is. Pitch Perfect re-initializes a custom persona at runtime using live research. That's the feature that makes it feel real.