Fig. 08 · Cold email with self-healing inboxes

BlazeMail.ai

Connects a user's own Gmail or IMAP inbox, personalises each email with an LLM, and runs multi-step campaigns under per-account sending limits, while keeping every mailbox connection healthy on its own.

Shut down
Role
Solo founder-engineer
Team
Solo, built at 17
When
2024 – 2026
Stack
Laravel 12, Inertia + React, Groq, Gmail API, IMAP/SMTP, Paddle, Pusher
Links
GitHub repo

01The problem

Founders doing cold outreach either write every email by hand or send obvious mail-merge. Sending at volume from your own inbox runs into rate limits, expired tokens and deliverability problems.

02How it works

6 steps

  1. 01 · System moves data

    Connect inbox and contacts

    Gmail OAuth or IMAP/SMTP, with contacts imported from CSV or Excel within plan limits.

  2. 02 · Rules decide

    Check credits

    Each prompt strategy (chain-of-thought, few-shot and others) has a credit cost checked before generation, and fraud heuristics flag rapid usage.

  3. 03 · AI proposes

    Write and personalise

    Llama 3 on Groq writes and personalises each email, falling back to a plain template merge if the API fails.

  4. 04 · Person approves

    The user reviews the batch

    Nothing sends until it's approved.

  5. 05 · Rules decide

    Throttled sending

    Per-account daily limits, delays between sends and campaign steps, and retry backoff. It rolls over to tomorrow when the limit is hit.

  6. 06 · System moves data

    Self-healing mailboxes

    Every two hours a health check finds risky accounts, refreshes tokens, and separates errors it can fix from ones that need the user to sign in again.

  • AI proposes
  • Rules decide
  • Person approves
  • System moves data

03Hard problems

3 of them
01

Mailboxes dying mid-campaign

Expired or revoked tokens and IMAP errors used to kill campaigns. The health loop refreshes what it can, marks the rest as needing re-auth, and tells the user.

02

Sending at volume without getting flagged

Daily limits are checked before and during every batch, with throttling, multi-step delays, retry backoff and inbox warm-up.

03

A campaign that survives the LLM going down

AI personalisation is optional per email and falls back to a deterministic merge on any provider error.

04By the numbers

active users on a freemium model
300+
mailbox health-check cycle
2 h
my age when I built it, solo
17

05What I'd do differently

Platform verification costs ended the business, and I only found out after building the product. Now I check platform compliance costs before I build on a platform.
Next · Fig. 01RAVNInfrastructure for physical AI