Instagram DM Intake
n8n intake layer for an AI appointment setter
Turns a burst of Instagram DMs into exactly one AI reply — verified, deduplicated, and handed back to a human when the coach steps in.
The problem
An AI setter that answers Instagram DMs needs a front door that answers Meta within a second, ignores retries and duplicates, waits for a lead to finish typing, and never talks over a human.
What I built
n8n workflows backed by Postgres that receive Meta's Instagram webhooks, verify every request, batch each conversation's messages, and pass one clean batch to a send sub-workflow. Built as a paid trial and delivered two days early.
Core features
- -Meta webhook verification handshake and HMAC-SHA256 signature check
- -Duplicate and retry protection through a database constraint
- -Burst batching: several quick messages produce exactly one reply
- -Automatic pause when the coach replies personally
- -Daily job that refreshes expiring access tokens
Engineering highlights
- -Signature verified over the raw request bytes with a constant-time comparison, so emoji and escapes never break it
- -A single SQL statement lets only the newest message claim the batch — exactly one reply even with concurrent executions, no locks or queues
- -Responds to Meta in about 40ms locally, before any database work
- -Signed-request test harness covering bursts, duplicates, parallel leads, emoji, and the coach-takeover case
Tech stack
n8nPostgreSQLMeta Graph APIWebhooksDockerBash
Impact
- -Client feedback called it "solid work", singling out the claim query and the limitations write-up