Skip to content
Augusta BhardwajSay hello
Work / 2026 · Paid ads

Admiral

AI for paid ads that never acts without approval

Audits Google Ads accounts, flags wasted spend, and proposes fixes the advertiser approves. Every change is logged, with undo wherever the ad platform allows it.

Admiralmeetadmiral.com/overview1 of 4
Admiral: Today. What needs a decision, the money behind each fix, and how the account is doing.
Admiral's console on its demo workspace, with example figures. Captured October 2026.
3,800+API Tests
Approve-firstChange Model
LoggedEvery Change
HourlyAccount Sync

The problem

Small advertisers burn budget on broad matches, wrong geos, and weak ads. Agencies are expensive, and the platform's own recommendations optimize for the platform. Running ads well takes daily attention most founders don't have.

How I built it

Built a web app that syncs account structure and performance hourly, runs AI audits that find wasted spend and missed opportunities, and ships every fix as a proposal the advertiser approves. Every applied change lands in an activity log as a receipt, with one-click undo wherever the ad platform allows a clean reverse. Infrastructure is Terraform-provisioned; the web app and API share generated types so the contract can't drift.

Next.jsFastAPITerraformGoogle Ads API

The hard parts

  1. Google Ads API depth: GAQL queries, mutate validation quirks, and rate limits across every entity type the audit touches
  2. Diff-and-undo semantics per change type, with undo offered only where the provider allows a clean reverse, and said plainly where it does not
  3. Autopilot guardrails: it is switched on by the advertiser, may only pause and block, and runs under a hard daily cap, so automation can never outrun trust
  4. Audit explanations an advertiser believes: every recommendation traces to the numbers that produced it