Crecer Digital

Crecer Digital — Personal AI Content Engine for a Solo Creator

Live demo: https://crecer-digital-content-engine.onrender.com Code: https://github.com/crecerdigital/crecer-digital-content-engine

Note on the demo mode, stated honestly: the public demo currently runs in LOCAL DEMO mode because no NEBIUS_API_KEY is set. The UI shows a LOCAL DEMO badge and the five agents run on the transparent local engine in engine.py. When a Nebius Token Factory key is configured, the Scriptwriter stage is enhanced live by an NVIDIA open model (Nemotron family) served by Nebius Token Factory, and the Analyst re-scores the improved script. Nothing is faked as live Nemotron output.

Q1 — What it does

Crecer Digital is a multi-agent content engine built for exactly one user: me — a solo Spanish-speaking creator in North Carolina whose wife and daughters live in Nicaragua. Five personal agents work as my always-on team: a Researcher agent turns a topic into script angles tuned to my three content pillars (faith growth stories, personal growth, prosperity principles); a Scriptwriter agent drafts 45-70 second vertical scripts in Spanish with the hook in the first three seconds; a Producer agent builds the shot list, on-screen text and caption plan; a Publisher agent prepares titles, descriptions and posting times for TikTok, YouTube and Facebook; and an Analyst agent scores every package against my quality rubric and flags what to fix. I approve every post from my phone before anything goes live — the agents produce, I decide.

Q2 — Who is it for? Why does it matter?

It targets solo and immigrant creators — the millions of one-person media businesses run by people who work a day job, speak to their audience in their own language, and have no team, no studio and no budget for either. For me it is personal: content is the bridge between the job I work alone in the US and the family I am working to reunite with in Nicaragua. A personal AI team that produces on schedule, in Spanish, at near-zero cost is the difference between posting daily and burning out — and everything it learns (topics, hooks, what my audience answers) stays mine, in my own memory store, not rented from a platform.

Q3 — What makes it different?

Three things. First, it is personal, not generic: the agents carry my pillars, my formats and my audience's questions, and they improve from my own performance data instead of giving everyone the same template. Second, it is a full chain, not a single prompt: research, script, production package, publishing kit and self-scoring happen as one pipeline with a quality gate — and a hard human-approval step before publishing, because my channels are my livelihood. Third, it is built for the real constraints of a solo creator: it runs on Nebius Token Factory with NVIDIA's open Nemotron models when a key is configured (and on a transparent local engine when it is not), costs cents per run, and works entirely in Spanish. The submitted repository contains the working MVP: the five-agent pipeline runs end-to-end in the browser, in the open.

Q4 — How did people test it? What was learned?

I am the user, and I test it every day: the engine's formats already run my real channels (daily faith stories, a Sunday long video, a prayer channel and a soccer channel). What I learned: (1) hooks decide everything — average watch time collapsed on videos that buried the point, and jumped when the story opened at its most emotional beat; (2) completion beats volume — one finished, well-scored video outperforms five rushed ones; (3) my audience answers comments far more when the video ends with a direct question about their own story. Those three lessons are now rules inside the agents: hook-first scripts, an Analyst score before anything is produced, and a comment-driving close in every single package.

Q5 — What's next?

Next, the engine runs its production fully on Nebius Token Factory with Nemotron models and a Tavily-powered research stage, and gains: (1) a persistent memory store so the Researcher and Analyst agents learn from every published video across months, not days; (2) a WhatsApp front door, so I can approve, edit or kill a post from the same chat where I talk to my family; (3) an audio-cleanup stage for the songs I record on my phone, feeding a second operation (original Spanish music with lyrics). The goal is a personal AI team that any solo creator could clone: your pillars, your language, your rules — running on open models, owned by you.

Built With

• nebius
• python
• render