Rumi — Preseed Investment Memo (Mar 2025)

Indexing all the world’s videos to enable real-time, AI‑powered content experiences.

By Escape Velocity · March 2025

Investment Overview

  • Decentralized video indexing network. Users (miners) run a browser extension while watching platforms like Netflix or YouTube. Edge ML produces context‑rich text annotations (not the videos themselves) that are sent to Rumi’s database for use by AI agents delivering contextual experiences, services, or ads.
  • Bootstrapping demand via an AI‑powered universal remote (UR) app. UR apps see ~4M monthly global downloads; Rumi’s first‑party app controls any smart TV and can act on voice prompts with context.
  • Proprietary audio fingerprinting. Identifies what a user is watching and their exact timestamp with <1s latency and >98% accuracy from a 400ms clip (ideal conditions), enabling real‑time agentic experiences.
  • Long‑term path to a “decentralized Nielsen.” Using DePIN playbooks to achieve scale and more statistically significant benchmarks than legacy panels, creating three‑sided effects between ad buyers, distributors, and agent developers.
  • Team & round. Serial founders with deep AI/consumer/crypto experience; EV3 is co‑leading the $4.7M pre‑seed with a16z CSX.

Network Overview

  1. Supply‑side miners: run the extension to expand coverage and depth of the annotations database, initialized from scripts/CC/translations. This compounds a catalogue moat over time.
  2. AI agent developers: leverage real‑time fingerprinting (what/where the user is watching) and the annotations DB for context. Rumi’s own UR app is the first agentic app; users can query in natural language (e.g., “rewind to the last first down”). Early usage earns inflationary token rewards.
  3. Demand‑side consumers: use apps powered by privacy‑preserving edge ML that isolates TV audio and runs locally, sharing only TV/video annotations.

The flywheel: a richer database → more powerful agentic experiences → better developer returns → more apps → more organic user queries that “fetch” missing context, further enriching the dataset.

WTNTB

Underwriting on a 6–10 year horizon:

  • ~250k developers paying ~$175/month for contextual DB access (implying a few hundred users per agent and a few cents per‑user per‑day).
  • Valuation framed vs. advertising data/analytics comps (historically acquired at ~6–11× EBITDA).

Key Risks

  1. Scope of indexing vs. disposable AI video. Value strongest for persistent content with durable annotations; TVs remain a large share of U.S. screen time.
  2. Bootstrapping demand. Strategy: first‑party UR app + 3rd‑party agent integrations (e.g., Virtuals, ai16z); UR category has strong organic demand.
  3. Content‑owner dynamics. Potential conflict on revenue capture; likely to be ignored early and partnered later if activity scales.
  4. Nielsen integrations moat. Must overcome via performant cross‑hardware architecture and DePIN‑style scale beyond ~50k homes.
  5. Competitive responses. Medium‑term tech advantage may compress; large incumbents (e.g., Amazon, Roku) are well‑positioned.
  6. Privacy & regulatory. Edge models and hashed 400ms clips limit leakage; fair‑use doctrine likely relevant given transformative text annotations.

References

Get the PDF version of the original investment memo:

Rumi — Preseed Investment Memo (Mar 2025) by Escape Velocity

📄 Download PDF