AI apps / Prospector
My own marketing research and outreach system.
Prospector finds businesses that look like a fit and drafts the first email. I approve every one before it sends.
What it does
Prospector finds businesses likely to need custom software work, scores how good a fit they are, and drafts a personal first email based on what it found. Nothing sends until I click approve.
How it works
- It researches businesses from a set of sources I’ve configured: search, public forums, job and tender boards, company sites.
- It profiles each company’s website, works out whether they’re a realistic fit, and writes down why.
- It drafts a first email that cites the specific facts it found about that business.
- The draft sits in a review queue. I read it, edit it or reject it. Nothing goes out on its own.
- Anything approved sends through my own mail server. Replies land in an inbox I read myself.
A person approves before anything goes out
This is the part I built first and won’t cut corners on. Every email Prospector drafts sits in an approval queue. I click approve, edit or reject, one at a time. Prospector never automates outreach on LinkedIn or Facebook either, it only ever drafts those messages for me to send by hand. Even when it rules a lead out, it writes down why, so a bad call is something I can catch, not something that happens silently.
Proof
Follow the work from research to a prepared email, then inspect the checks that can hold it for review. These mockups use fictional businesses and messages.
Pipelines. From research to a reviewed message: 25 process areas across five workflow groups, with checks and a person’s approval along the way.
Governance. Eight review controls make the reason for a hold visible. Here, an unsupported claim needs revision before the draft can move forward.
Email review. The prepared email sits beside the evidence behind it. A person can edit, reject or approve the draft before anything is sent.
Reconstructed mockups with fictional data. The workflow groups summarise capabilities; they do not reveal internal pipeline definitions, prompts or decision rules.
For your technical person
Prospector is hosted and run on local or in-house physical servers. Its browser-based interface, research workers and review queue run within that deployment.
Built with Next.js, React, TypeScript and Node.js, it stores research records and review history in a database. Research, drafting and email delivery run as separate background workflows, with public-source research and email integrations feeding the review queue.
Source evidence, draft revisions and review decisions stay linked to the work. A person can inspect the supporting facts, revise or reject a draft, and approve each outgoing message before it is sent.
Related help.
Tell me who you’re trying to reach. I’ll tell you what Prospector would do.
+27 87 150 9305Monday to Friday, 09:00 to 20:00 SAST. Or email info@antondevilliers.com.