Experience
–
AI Engineering Intern
MAKKN Egypt
- Found that RNNoise denoising was degrading ASR for an Egyptian Arabic voice agent and bypassed it on the STT path, cutting WER from 54.8% to 29.6% (60 clips including 8 kHz telephony).
- Benchmarked five open-source Egyptian and code-switching ASR models at 16 kHz and simulated 8 kHz telephony, and wrote the STT, LLM and tashkeel selection decision records.
- Built evaluation harnesses for ASR, LLM, TTS and code-switching: LLM-as-judge, an end-of-turn benchmark and live-audio re-scoring.
- Built a dedicated vLLM ASR service and shipped RAG, tool calling and latency instrumentation on a four-person team (40 of 63 merged commits).
–
Network Infrastructure, AI & Automation Intern
Nokia Egypt
- Built a RAG assistant over a 1,568-page Nokia hardware planning guide (structure-aware chunking, manual cosine search, cite-or-refuse prompt) that scored 8/8 on a fixed evaluation, including a correct refusal.
- Built a four-screen Power Apps task manager on SharePoint with a Power Automate email-notification flow as the program capstone.
–
Cloud Engineering Intern
Obelion.ai Remote
- Designed how a cloud-provisioning platform's Ansible engine would run on Kubernetes (Jobs, RBAC, Secrets, EFS volumes, templated manifests) and prototyped job creation with the Node client.
- Built an HA RabbitMQ cluster with the operator and Kustomize overlays, plus Terraform modules for RDS and Amazon MQ.
- Built a GitHub Actions to S3 to Auto Scaling deployment with a least-privilege IAM user, EFS and a private database behind a bastion TCP proxy, and benchmarked PostgreSQL dump formats on 1.5 GB (parallel restore about 3x faster).
–
Backend Engineering Intern
3D|Diagnostix Remote
- Analyzed a legacy Node.js order controller (22 endpoints), found contradictory billing logic and redundant queries, and wrote endpoint-level test plans and logging improvements.
- Designed and demoed a performance-monitoring tool (response time, TTFB, percentiles) with a 51-hour plan for dashboards and regional probes, and hypothesised the data layer as the main bottleneck (to be validated).
–
DevOps Intern
eSpace Egypt
- Built an AWS staging environment (VPC, bastion nginx reverse proxy with HTTPS and basic auth, private Jenkins host) used by software, QA and pentest interns.
- Delivered Jenkins CI/CD pipelines for the three apps of an internal AI product (React, Rails, Python): Docker Compose builds from the staging branch, health checks and per-build release directories.
- Wrote Terraform for a three-AZ AWS network (S3 remote state, bastion, RDS subnets, security groups, modules) and Ansible roles for LAMP and LEMP stacks.
- Diagnosed recurring staging outages from system and application logs (multi-gigabyte AI Docker images from Jenkins builds filling the disk) and added scheduled image cleanup; no further outages.