CRM and agent workspace
Contacts, pipeline, tasks and call history in one place, built into the call workflow.
Prime Alley
AI communications platform
Apex, Prime Alley's AI communications platform: AI that answers and places calls, follows up on WhatsApp and hands off to people, with a CRM and agent workspace built in.
130–140
Concurrent calls handled through the working day
3
AI models on every call recording
Zero
Downtime during deployments
- Client
- Prime Alley
- Sector
- Managed IT and AI communications, UAE
- Service
- AI Automation
About the project
Apex is Prime Alley's AI communications platform. It answers inbound calls, places outbound calls, follows up over WhatsApp, email and meetings, keeps contact records up to date, and hands conversations to human agents with the transcript and summary attached.
Agents work from a dashboard of contacts, pipeline, tasks and meetings, while supervisors get a live console and analytics on call outcomes. Because call recordings must stay under the company's control, the whole platform runs self-hosted on its own machines.
Scope
What our team worked on
WhatsApp follow-ups
Automated follow-up flows that pick up where a call leaves off.
AI call pipeline
Transcription, translation and analysis on every call recording.
AI co-pilot
A self-hosted LLM agent that can operate the platform through conversation.
Infrastructure and releases
Self-hosted deployment on the company's own machines, with zero-downtime releases.
Highlights
Problems we solved
01 Recordings couldn't leave the company
Compliance rules meant call recordings had to stay on infrastructure the company controls, ruling out public cloud AI services.
How we solved it
Ran the entire platform, including three AI models over every recording (transcription, translation and analysis), on Prime Alley's own on-prem machines.
02 Releasing without dropping live calls
The platform handles calls all through the working day, so a normal deploy would cut off live conversations.
How we solved it
Rolling releases across two machines: GitHub Actions builds images to GHCR, traffic shifts to one machine while the other updates, then load balancing resumes.
03 Running the platform by talking to it
Users wanted to manage the platform without clicking through screens.
How we solved it
Built a self-hosted Llama agent with tool calling that can operate the whole app, with most of the effort spent making its actions reliable and safe.
Outcomes
Results
- Handles roughly 130–140 concurrent calls through the 11:00–20:00 UAE working day
- Call recordings and AI processing stay in-house, meeting the compliance requirement
- Releases ship without interrupting live calls
- Client accounts are being migrated onto the platform, with more in testing
Tech stack
- Python
- FastAPI
- PostgreSQL
- Llama
- Docker
- Nginx
- GitHub Actions
Have a project like one of these?
Tell us what you're working on. We'll reply within one business day with next steps and a clear plan.