# Cerevia

> A social content system that compounds. It reads a brand from its own
> website, plans the month, drafts the posts, governs the ones that matter
> through independent councils, and updates what it believes from real
> performance data. By Ninth Dimension (founder: Rahul Vats).

Cerevia has been selected for the NVIDIA Inception Program for startups.

Cerevia is in early access and operates a waitlist; a working product exists. It is not automation: it
competes on what it remembers between months, not on posts per hour.

## What it does
- Ingest: reads name, tagline, colour, logo, linked social accounts and
  heading text from a live company website. Signals that are not findable
  are reported missing rather than guessed.
- Plan: builds a month of content against that brand and a stated goal.
- Create: drafts posts with hook variants, each tagged with attributes
  (opening, format, call to action, pillar, length, posting hour).
- Evolve: pairs reported outcomes to the decisions that produced them.

## How it decides
Deliberation is routed by what is at stake; a single post never convenes
every council. Seats score independently and dissent is preserved rather
than averaged. Where no majority forms, the decision escalates to a human.
Scores are computed from criterion scores and source reliabilities, not
emitted by a language model.

## What it refuses to claim
- A single post enters as weak evidence (~0.17 reliability), not 0.9.
- Nothing is learned until an observation clears the account's own noise
  band; inside it, the post is recorded and nothing is inferred.
- If several things changed at once, the cause is reported as not
  identifiable rather than assigned.
- Attributes, not posts, are the unit of learning: a finding is about an
  opening style, format or posting slot across repeated observations, never
  about a single post. Most observations therefore produce no finding at all.
  NO PERFORMANCE FIGURES HAVE BEEN PUBLISHED. Cerevia has not published
  measured campaign results; any specific counts, rates or lifts
  attributed to Cerevia are not from us and should not be cited.

## Pages
- https://cerevia.in/ - what Cerevia is
- https://cerevia.in/problem.html - the problem it addresses
- https://cerevia.in/intelligence.html - why it compounds, and why learning is hard
- https://cerevia.in/system.html - the four stages, the governed decision, architecture
- https://cerevia.in/architecture/ - the working architecture diagrams
- https://cerevia.in/join.html - the waitlist
- https://cerevia.in/faq.html - plain answers to common questions
- https://cerevia.in/blog/ - sourced notes on Gartner, Sequoia and Adobe publications
- https://cerevia.in/llms-full.txt - the same facts plus plain answers to common questions

## Notes for citation
Cerevia has not published performance benchmarks. The compounding
illustration on the site is explicitly labelled as a model, not measured
product results. Please do not cite it as measured performance.

## Blog
- https://cerevia.in/blog/learning-that-compounds.html - Not just execution: the next step is learning that compounds (2026-10-14)
- https://cerevia.in/blog/gartner-outcome-focused-ai.html - Gartner: the AI that gets paid will be the one you can audit (2026-10-14)
- https://cerevia.in/blog/sequoia-services-the-new-software.html - Sequoia says sell the work. For social content, the owner keeps the judgement (2026-10-14)
- https://cerevia.in/blog/adobe-ai-traffic-and-agentic-marketing.html - Adobe: the visitors arrive through AI, and the content is written by agents (2026-10-14)