Sparq Intelligence
A lead-research platform that labels every AI finding by its evidence and drafts outreach for a salesperson to approve
- Type
- AI platform
- Status
- Private system
- Year
- 2026
- 11 steps from business search to CRM follow-up
- 3 evidence labels: observed, inferred, hypothesis
- 512 test methods
Sparq Intelligence is an internal tool for Sparq Web. It finds local businesses with unmet digital, software, automation and AI needs, works out what is wrong with their current presence, maps that to a service, and drafts a personal email that a salesperson approves before it goes out.
Most of the work went into keeping the output honest. The system separates what it observed on the business’s website and reviews from what it inferred and what is only a hypothesis, and shows that label next to every finding a salesperson reads. Every finding has to name its evidence, and a person reads each draft before it sends.
The platform also had to run on ordinary shared hosting, driven by a single cron entry. So every stage is its own small queued job: a worker started by cron drains the queue in 50-second runs, and the next run picks up where the last one stopped.
How it works
- Business search
- Business profile
- Website check
- Review check
- AI analysis
- Opportunity detection
- Lead score
- Personalized outreach
- Human approval
- Send email
- CRM follow-up
Screens
Engineering decisions
- Every AI finding carries evidence, a confidence level and a label: observed, inferred or hypothesis. Inference is never presented as fact.
- Capabilities that can't be verified from public sources stay "unknown" instead of being guessed.
- No outreach email leaves the system until a salesperson approves the draft.
- It runs on ordinary shared hosting with no Docker, Redis, queue daemon or Node server in production, and deploys from GitHub on every push to main.
Built with
- Laravel
- PHP
- Claude API
- MySQL
- Alpine.js
- GitHub Actions