A secure, high-volume platform
I own the design and delivery of a secure, high-volume client platform, side by side with the client's infrastructure, security and compliance teams.
6 engineers · backend, frontend, DevOpsMohammed Alostah here. I lead engineering at Tarjama Group in Amman, where my team and I run high-availability platforms on Azure Kubernetes for enterprise and government clients across the GCC and Levant. I've spent 7+ years on backend and distributed systems, and now I put LLM agents into production where they save real hours.

Design reviews, hiring, rollout plans and the pager at 3 a.m. Here's what came out the other end.
I own the design and delivery of a secure, high-volume client platform, side by side with the client's infrastructure, security and compliance teams.
6 engineers · backend, frontend, DevOpsA LangChain and CrewAI pipeline that took the repetitive steps out of everyday jobs, so people could focus on the parts that need judgment.
2M+ words/month · −40% manual work · 60% fasterA multi-tenant Laravel and Node.js backend on AKS, with Redis queues and auto-scaling taking the load.
10k+ users · 500k+ API calls/moBroke a monolith into containerised services, so the team went from biweekly releases to several a week.
Deploys: 45 min → under 10Threat models with STRIDE, Azure AD single sign-on, F5 WAF policies, VAPT remediation and data-sovereignty controls.
Passed client security reviewsOpen-source experiments where I try out agents, small language models, speech and search.
An AI-native, multi-tenant contact center. Voice, chat and automation all enter one durable agent runtime with RAG, tools and memory, handing off to humans when it should.
A local coding agent CLI on Google ADK and LM Studio. It routes each request, then runs an explore, plan, code and validate loop.
A small open language model family built from scratch, from a single trainable weight up to a 45M-parameter GPT, with weights on Hugging Face.
A speech recognition service that also tells speakers apart and detects the language, built on OpenAI Whisper and NeMo.
A Laravel client for the Vespa search engine, with a query builder, rate limiting, caching and plugins.
An AI-powered web crawler that extracts text from sites and analyses the content.
Al-Zaytoonah University of Jordan, 2016
Hiring an engineering lead, or want a second opinion on your architecture or AI plans? Tell me what you're working on.
eloustah@gmail.com