AI sales platform for B2B
2025 – 2026RoleFull-stack developer
A SaaS where businesses set up AI sales agents for calls, SMS, email, messengers and a web widget.
- Laravel
- Vue
- FastAPI
- PostgreSQL
- Redis
- Docker
Solution
As a full-stack developer in a large team I owned much of the customer dashboard and several back-end services:
- Lead distribution: round-robin, weighted, team and hybrid strategies with manager availability and calendars; a drag-and-drop manual order.
- Calendars and booking: Google Calendar and Calendly with webhooks, two-way sync, a free-slot service across all calendars, protection against double booking, booking exposed to the AI agent through function calling.
- AI agent: function calls for slots, booking and contact updates; knowledge-base search with incremental indexing; language switching with confirmation; six conversation styles; guardrails against prompt injection; token checks before every model call.
- Flow builder: a redesigned visual editor with variables, validated edges and auto-layout, autosave and rollback, and a migration of all flows from JSON to relational tables without data loss.
- Channels: one configurator for voice, email, SMS and iMessage with real-time validation and a channel-integrations page.
- Telephony: call history, recordings and transcripts, voice settings for several speech providers, call campaigns from CSV, browser test calls, a move to streaming speech recognition, and a SIP bridge with two carriers — including a race condition between carrier events and SIP signalling that I tracked down and fixed.
- Token billing: a balance dashboard with forecast, Stripe pay-as-you-go, subscriptions, auto top-up and invoices, live balance over WebSocket, plan limits and discounts all the way to Stripe, unlimited owner roles synced between two databases.
- Phone-number marketplace: number search and purchase priced in tokens, regulatory verification by country, spending reports, audit log and roles.
- Security: two-factor login including after OAuth, SAML SSO, service-to-service JWT, PII masking.
- DevOps: Dockerfiles for every service, migration of production from AWS to Azure with infrastructure as code, runbooks and an atomic switch of 13 DNS records; PgBouncer for 12 Python services after connection-pool exhaustion.
- Quality: about 25,000 lines of tests (PHPUnit, Vitest, pytest, E2E).