GTM engineering
The Lead Machine
The entire outbound sales motion as one system: source leads from the open web and data providers, tier them against the ICP, enrich, push to HubSpot, and route each tier to the right channel, whether that is LinkedIn, email, or another motion entirely. It drafts signal-based messages per persona, runs the sequences, tracks who accepted and who replied, and logs every step. One operator doing what used to be a full SDR function.
I built this demo as one self-contained page: no video, no backend. Companies, contacts and numbers are fictional; the mechanics are real.
Best viewed on a laptop: Open the full demo
Context
An early-stage startup can't hire an SDR team. It can build one. The job was every step of outbound, not just the top of the funnel: find companies, decide if they fit, find the right person, get their email and LinkedIn, connect, write something worth replying to, sequence it, and know which of it worked.
What I built
- ICP as an executable YAML rulebook, not a slide
- A permanent registry: a repeat search over the same sources yields zero new rows
- Nine numbered Python scripts underneath, no LLM SDK inside
- Eight skills on top: enrich, sync HubSpot, connect on LinkedIn, audit accepts, draft, sequence
- Schema and field mapping defined across intake, qualification and CRM; repeated imports and failure cases verified
- A prioritized backlog with ticket status, delivery checks and a completed-work archive; every process documented for the next run
How it runs
A GitHub Actions job re-checks outcomes every 6 hours and commits data changes back. The source scorecard is the actual decision tool: every channel is ranked by its measured reply rate and budget follows the numbers. A/B tests on email copy run with real confidence intervals, and a copy defect found in production (a dangling reference in one batch of sent emails) became a permanent rule in the drafting skill.
What it taught me
The bottleneck was never volume. It was knowing which of six sources to trust and which of twelve message variants to keep. A scorecard beats an opinion every time a channel gets defended in a meeting.
Stack
- Python
- YAML config as rulebook
- HubSpot
- MeetAlfred
- Instantly
- CoreSignal
- Apollo
- Luma
- GitHub Actions
- agent skills
Want the walkthrough?
The code is private. The walkthrough isn't.
Book 30 minoranov.con@gmail.com