# Axelerated Solutions: Lead Backend Engineer

Source: https://www.alouaniyoussef.com/experience/axelerated

- Period: Feb 2024 – May 2026
- Location: Remote
- Founding engineer
- Team size: 9

Founding engineer and backend lead for Sowtek, an omnichannel customer-engagement platform, with speech-to-text, RAG and MCP at its core.

## Context

Axelerated Solutions builds Sowtek, an omnichannel customer-engagement platform for contact centres in Saudi Arabia: calls, WhatsApp, email and social media in one place, with AI on top. Its clients include government and security organisations, so privacy and deployment flexibility matter as much as features.

## My part

I was a founding engineer and the backend lead in a team of nine.

### AI systems

- Fine-tuned Whisper large-v3 for Saudi-dialect call transcription, with an audio pipeline for WhatsApp and Asterisk calls.
- Designed the Customer Voice RAG pipeline on Milvus, with an MCP server as its main retriever, and owned its architecture, auth, privacy and retention.
- Built an MCP server over our backend APIs, with every request validated through Keycloak.
- Built coding agents, tool-using chatbots, and knowledge-base search with Docling and zero-shot classification.

### Contact-centre backend

- Omnichannel chat: WhatsApp, email, X, Instagram and Facebook behind one conversation model.
- VoIP on Asterisk: PBX event processing, call-recording pipelines and sentiment on transcribed calls.
- FlowBuilder: the engine behind Sowtek's low-code call, chat and SMS workflows.

### Migrations and deployments

- Business model: B2C to B2B.
- Architecture: stateful to stateless services.
- Orchestration: Docker Compose to Kubernetes.
- Hosting: Hetzner to Oracle Cloud.
- Hybrid deployments for private, banking and government institutions.

## Stack

NestJS, Python, FastAPI, gRPC, Milvus, RAG, MCP, Whisper, Fine-tuning, Docling, Zero-shot classification, Keycloak, Kubernetes, Ansible

## Outcome

Shipped to government and enterprise clients, including HRSD and Tahreez. Speech-to-text and Customer Voice figures are on their case studies.

## What I'd do differently

I would have built the evaluation set for Saudi-dialect speech before the first fine-tuning run, rather than alongside it.

## Clients

### Government and public institutions

- HRSD (Ministry of Human Resources and Social Development)
- King Saud Medical City

### Growth and technology companies

- Tahreez
