How much does it cost to run?

About $3/month at typical small-business volume (around eight long pieces a month, each fanned out into a batch of short drafts). The fixed cost is essentially zero. The variable cost is dominated by the Bedrock calls that pull points and draft each clip; everything else (Lambda, DynamoDB, S3, scheduling) is pennies. At a heavy 40 pieces a month the bill lands around $12.

Which AWS services does it use?

Lambda (Python 3.14, arm64) with Function URLs for the approve/edit/skip buttons, EventBridge Scheduler for the weekly drip and one-off jobs, DynamoDB on-demand, S3 (with versioning), SES outbound, Secrets Manager, CloudWatch Logs (7-day retention), AWS Budgets, and Bedrock (Claude Haiku 4.5 to pull points and draft, Claude Sonnet 4.6 only for the harder long pieces) via Global cross-Region inference, with Titan Text Embeddings V2 and S3 Vectors so each draft can quote the exact passage it came from. No API Gateway, no NAT Gateway, no always-on compute.

Where does the source come from?

Three lanes. Drop a blog post or doc in a Google Drive folder you control. Forward a recording transcript to a dedicated address. Or paste a link to a published post. A small sync Lambda mirrors the source to S3 so the drafter reads from there and gets S3 versioning for free. The Drive folder stays the place you control the source.

Does it use AI, and can it make things up?

It uses Bedrock to pull the strongest points and to draft each short post, but every draft is grounded in the original. The drafter is only allowed to use sentences it can trace back to a passage in the source — each draft ships with the quote it came from, so you can check it in one glance. If a claim can't be traced to the source, it's dropped. The model rewrites for length and platform; it doesn't invent facts.

Does anything post on its own?

No. Every draft waits for you. Each one has three buttons: Approve (send it to your scheduler or your channel), Edit (open a box pre-filled with the draft to tweak it), and Skip (drop this one). Nothing leaves the review desk without a human tap. Approved drafts can drip out over the week via a scheduler, but the approval itself is always yours.

Series · 7 parts Published May 26, 2026

Content repurposer

A serverless system that turns one long piece — a blog post or a recording transcript — into a week’s worth of short posts. It reads the source, pulls the strongest points, and drafts platform-sized versions in your voice: a thread, a few short posts, a quote-card caption. Every draft is grounded in the original, so it doesn’t make things up. You approve, edit, or skip each one; nothing posts on its own. Seven posts on the same system — one diagram at a time — with an engineering reference at the end.

01

A content repurposer on AWS for a few dollars a month

The whole system on one page — a source intake, a drafter, and a review desk, plus the rule that keeps every short post traceable back to the original.
02

How a long post gets loaded

Three lanes feed the source store — a Drive folder you drop posts into, a forwarding lane for recording transcripts, and a paste-a-link lane that fetches a published post and cleans it down to plain text.
03

How the strongest points get pulled

The source is split into passages, each one indexed, and Bedrock picks the handful worth posting — every chosen point carries the exact passage it came from, so nothing is invented.
04

How a clip gets drafted

Each chosen point becomes platform-sized drafts in your voice — a thread, short posts, a quote-card caption — written from a voice doc and checked against the source before it reaches you.
05

How a repurposed draft gets approved

Three buttons on every draft: approve (send it on or queue it), edit (tweak it in a pre-filled box), and skip (drop it). Nothing posts on its own, and every choice is logged.
06

What the content repurposer costs

A few dollars a month at SMB volume. The system only runs when you give it a piece, and the cost is dominated by the Bedrock calls that pull points and draft each clip.
07

Engineering reference: the content repurposer architecture

Same system, drawn purely for engineers. Service names, resource identifiers, region, Bedrock model IDs, the S3 Vectors index, Lambda inventory, IAM scopes, EventBridge Scheduler config, and the DynamoDB schemas.

What is a content repurposer?: A small serverless system that turns one long piece — a blog post or a recording transcript — into a week’s worth of short posts. It reads the source, pulls the strongest points, and drafts platform-sized versions in your voice: a thread, a few short posts, and a quote-card caption. Every draft is grounded in the original so it doesn’t make things up. You approve, edit, or skip each one; nothing posts on its own.
How much does it cost to run?: About $3/month at typical small-business volume (around eight long pieces a month, each fanned out into a batch of short drafts). The fixed cost is essentially zero. The variable cost is dominated by the Bedrock calls that pull points and draft each clip; everything else (Lambda, DynamoDB, S3, scheduling) is pennies. At a heavy 40 pieces a month the bill lands around $12.
Which AWS services does it use?: Lambda (Python 3.14, arm64) with Function URLs for the approve/edit/skip buttons, EventBridge Scheduler for the weekly drip and one-off jobs, DynamoDB on-demand, S3 (with versioning), SES outbound, Secrets Manager, CloudWatch Logs (7-day retention), AWS Budgets, and Bedrock (Claude Haiku 4.5 to pull points and draft, Claude Sonnet 4.6 only for the harder long pieces) via Global cross-Region inference, with Titan Text Embeddings V2 and S3 Vectors so each draft can quote the exact passage it came from. No API Gateway, no NAT Gateway, no always-on compute.
Where does the source come from?: Three lanes. Drop a blog post or doc in a Google Drive folder. Forward a recording transcript to a dedicated address. Or paste a link to a published post. A small source-sync Lambda mirrors the source to S3 so the drafter reads from there and gets S3 versioning for free. The Drive folder stays the place you control the source.
Does it use AI, and can it make things up?: It uses Bedrock to pull the strongest points and to draft each short post, but every draft is grounded in the original. The drafter is only allowed to use sentences it can trace back to a passage in the source — each draft ships with the quote it came from, so you can check it in one glance. If a claim can’t be traced to the source, it’s dropped. The model rewrites for length and platform; it doesn’t invent facts.
Does anything post on its own?: No. Every draft waits for you. Each one has three buttons: Approve (send it to your scheduler or your channel), Edit (open a box pre-filled with the draft to tweak it), and Skip (drop this one). Nothing leaves the review desk without a human tap. Approved drafts can drip out over the week via a scheduler, but the approval itself is always yours.

All posts

Frequently asked questions

Other series