Guglielmo Paesano AI engineer and founder of AIGROW

I build AI systems that do people's work, and I measure where they still need people.

contact@guglielmo.ioDiscord, 4K members

ReelVance

An autonomous short-video engine that ran five accounts daily with no human in the loop.

ReelVance was an autonomous short-video engine. It found what was trending, wrote the script, rendered the video with synthetic voice and captions, and published to TikTok, Instagram and YouTube by itself. It ran in production from July to November 2026 and continues as an in-house tool.

Status
Archived
Role
Founder and lead engineer
Year
2026
ReelVance: reelvance.com, filmed as it runs

What it is

An autonomous short-video engine, built solo and run in production from July to November 2026. It found what was trending, wrote the script, rendered the video with synthetic voice and captions, and published to TikTok, Instagram and YouTube by itself. Archived as a product in November 2026 to put full focus on AIGROW; the engine still runs as an in-house tool.

Built with

  • TypeScript
  • Remotion
  • Next.js
  • Stripe
  • Playwright
  • MCP

What you can check

  • Ran five accounts every day with nobody approving each step
  • First-party OAuth publishing to Google, TikTok and Meta
  • An append-only credit ledger behind the paid console

The problem it solved

Short-form video rewards volume and consistency, and both break on the same bottleneck: a person has to decide what to make, make it and post it, every day. Editing tools and schedulers each remove one of those steps. ReelVance was built to run all of them unattended.

A creator using an editor and a scheduler still sits down every day to a blank prompt, because neither tool decides what to publish.

The engine was built so that a person changed the strategy and the daily execution ran without them. That is a harder system to build, because every step has to be safe to run with nobody watching.

How the chain worked, end to end

Four stages ran in sequence on a schedule: find a subject that is trending, write the script, render the video, publish it. Each stage wrote its output for the next, and each could fail without taking the chain down, because a stalled stage can be retried and a bad publish cannot be taken back.

Rendering was Remotion with 3D avatars, synthetic voice, automatic captions and beat-synced editing. Scripts came from a language model, with the choice of model kept as configuration.

Publishing was first-party OAuth with Google, TikTok and Meta, with Playwright and MCP agents for the steps no API covers. Building those integrations took weeks, and it is the reason the engine could run unattended: a posting service that rate-limits you or loses its token is a dependency you cannot debug.

A ledger where a counter would have been

Usage was recorded as an append-only ledger of entries. A counter says a customer has 40 credits left. A ledger also says what happened to the other 20, which is the question that gets asked when something goes wrong in a system that spends on its own.

An autonomous system spends without asking each time. The first time a run misfires, a balance number leaves nothing to reconstruct. With a ledger, every deduction is attributable to the run that caused it, and a support conversation becomes a lookup.

Publishing with nobody watching

The engine ran five accounts every day with nobody approving each step. Every video carried a safety report scoring originality, AI disclosure and policy risk before it went out, because one post that breaks a platform policy can cost an account, and no retry recovers that.

A pipeline that runs for a week looks finished. One that runs for months meets token expiry, platform API changes, a model that starts refusing a prompt it used to accept, and a render that silently produces a black frame. Each of those became a check.

The lesson carries into every agent I build for a client: a system that acts on its own needs a gate that is not a person, and a record of what it did that someone can read afterwards.

AI agent development, as a service

Why it is archived

ReelVance was archived as a product in November 2026 to put full focus on AIGROW. The engine still runs as an in-house tool. The paid console, with its Stripe billing, credit ledger and usage limits, is no longer open to new customers.

It is listed here because the engineering is the same kind a client hires me for: a chain of model calls and tools that has to do a whole job, every day, without supervision.

Questions

What did "autonomous" mean here?
That the chain from choosing a subject to publishing the finished video ran on a schedule with nobody approving each step. A person changed the strategy, the prompts and the accounts. The daily execution was not a person's job.
Why build first-party OAuth publishing instead of using a posting service?
Because an intermediary is a dependency you cannot debug. When a third-party service rate-limits you or loses a token, the failure surfaces on your account and the fix is not in your hands. For a system designed to run unattended, that is the one class of dependency worth the weeks it costs to remove.
How was credit usage tracked?
As an append-only ledger in which every deduction is attributable to the run that caused it. A balance that only goes up and down cannot explain itself after a bad run, and a ledger can.
Is ReelVance still available?
No. It was archived as a product in November 2026 so that AIGROW could have full focus. The engine still runs as an in-house tool.

Facts on this page verified 2026-11-10

Tell me what you need built.

Write what the system has to do, what exists already and when it has to be live. I read every brief myself and answer in writing, with a scope and a price for each milestone.

Or write to contact@guglielmo.io

The human check appears here.

No calls and no sales sequence. One written answer from me.