One outbound AI call becomes structured, actionable intelligence without a human reviewing the transcript.

1. 1A call is placed — single, batch or scheduled — with per-prospect variables handed to the ElevenLabs Conversational-AI agent
2. 2The prospect flips to calling and a call record is opened; the scheduler later syncs the finished conversation back and hydrates the transcript
3. 3OpenRouter analyses the transcript into sentiment, intent, interest score, key points, objections and extracted contact fields
4. 4The insight cascades: it updates the prospect’s stage, schedules a callback, raises ranked action items and logs consent
5. 5A compliant follow-up email is drafted from the call, ready to review or auto-send from an encrypted mailbox

An anonymous website visit is turned into an enriched, CRM-ready lead in the background.

1. 1A cookieless beacon registers the session and streams batched pageviews, clicks, scroll depth and form events
2. 2The server derives a cross-session fingerprint from request signals to re-identify returning visitors
3. 3Each new visitor is enriched with IP geolocation, ISP/VPN detection and company/email identity, with graceful fallback when a provider is unavailable
4. 4Click coordinates are aggregated server-side into per-URL heatmaps for the analytics admin
5. 5A submitted contact or package request is captured as a lead and fanned out to the CRM, with each delivery flagged idempotently

A dedicated background service keeps the whole funnel moving without anyone pressing a button.

1. 1A separate scheduler container runs a short-interval loop alongside the API
2. 2It auto-syncs calls per user, fires due callbacks, and reconciles ElevenLabs campaign batches
3. 3Due callbacks are claimed with a conditional update before dialing, so overlapping ticks can never double-dial
4. 4It drains the email outbox, polls IMAP for replies, and materialises recurring call campaigns on schedule
5. 5Failed calls, campaigns and runs are retried with exponential backoff and separated from terminal failures for manual review

A push to main ships either system independently, health-gated, with no manual steps.

1. 1A merge to main triggers that system’s GitLab pipeline
2. 2CI rsyncs the code to its server, excluding data, media and secrets to protect persistent state
3. 3It rebuilds the Docker image and brings the compose stack up — the sales system as a web plus scheduler pair
4. 4Migrations and static collection run inside the container, then the health check is polled with retries before traffic resumes
5. 5The two systems deploy to separate hosts (DigitalOcean and Hostinger) with their own pipelines, while the frontends ship to Vercel and Netlify