• Maintain and recover EC2, EBS, and S3 host fleets, including P4d and P5 GPU capacity used for large-scale ML training, returning hardware to service against strict availability targets.
• Install, configure, and troubleshoot compute and network racks across the physical compute, storage, and networking layers that EC2, EBS, S3, and VPC are built on.
• Drove on-site diagnosis of an inter-site DWDM metro span failure that removed link redundancy between two data centers, coordinating with optical on-call engineers, the fiber vendor, and the colocation provider until the fault domain was isolated and redundancy restored.
• Lead operational and safety training for a six-person technician team and produce the site's monthly operations reporting.
• Contribute to post-incident reviews with network engineering and vendor teams, feeding findings back to the service teams that own the affected systems.

• Selected for a competitive 23-week full-stack fellowship covering JavaScript, React.js, RESTful APIs, PostgreSQL, and HTML/CSS.
• Built and deployed multi-tier web applications, including an e-commerce platform with a relational database backend, across frontend, backend, and data layers.
• Delivered projects against weekly milestones with code review and stakeholder demos.

• Held 100% uptime across all critical electrical and precision cooling systems (UPS, PDU, CRAC, feeders, transformers) in a Tier III data center serving financial-services clients.
• Racked, installed, configured, and troubleshot 100+ servers, storage arrays, and network devices.
• Ran incident and problem management in Remedy (ITSM), including structured root cause analysis and change control.
• Produced daily, weekly, and monthly availability reporting against client SLAs.
• Supervised facility and IT vendors on site, enforcing documented procedures and operational checklists.
• Produced audit-ready documentation for internal reviews and external compliance assessments.

• Audited national frequency allocations and proposed a reallocation that was adopted, recovering usable spectrum from idle assignments.
• Authored technical reports used in regulatory decisions on mobile data service quality.
• Worked with cross-border counterparts on frequency-sharing agreements.

AI Model CI/CD Pipeline (MLOps)

An MLOps CI/CD pipeline for deploying LLM models on AWS EKS with automated canary deployments and rollback. Three-stage workflow: CI (lint, test, model evaluation against quality thresholds), staging deployment with smoke tests, and production canary rollout with Prometheus-based health monitoring. Terraform provisions ECR, S3, and IAM with GitHub OIDC for keyless auth; Kustomize handles environment overlays; GitHub Actions orchestrates. A model evaluation framework gates releases on p95 latency, throughput, quality score, and error rate.

Mistral RAG Assistant

A retrieval-augmented generation application powered by Mistral AI, with semantic search over Qdrant, multi-format document ingestion (PDF, DOCX, TXT, MD), recursive chunking with overlap, streaming chat with source citations, and an evaluation framework measuring retrieval precision, answer relevance, faithfulness, and correctness. Containerised with Docker Compose and deployed to AWS ECS via GitHub Actions.

Mistral LLM Inference Platform on AWS EKS

An LLM inference platform serving Mistral 7B Instruct on AWS EKS with vLLM on NVIDIA A10G GPUs (Spot instances for 60–70% cost savings). A FastAPI gateway provides bearer-token auth and per-IP rate limiting; autoscaling runs on a custom HPA metric (pending inference queue depth); Terraform provisions VPC, EKS, ECR, and managed node groups; Helm handles deployment; Prometheus and Grafana provide observability; CI/CD implements canary rollout with automated smoke tests and rollback. Load-tested with Locust at a 0% error rate and ~3.4 s p50 latency for 100-token completions.

Secure Multi-Account AWS Landing Zone for Banking

An AWS Landing Zone for French banking institutions subject to DORA and GDPR. Seven dedicated accounts across four OUs, centralised security monitoring, immutable audit logging (365-day S3 Object Lock), hub-and-spoke networking via Transit Gateway with route table segmentation, dual-path hybrid connectivity (Direct Connect plus VPN), and cross-region disaster recovery designed for RTO 4 h / RPO 1 h. Six Service Control Policies enforce region restriction, encryption, and service allowlisting. Codified in 10 reusable Terraform modules with a simulation mode for single-account testing.

AWS Cloud Resume Challenge

A full-stack serverless website on S3 (static hosting), CloudFront (CDN with ACM/HTTPS), Route 53 (DNS), DynamoDB (visitor counter), Lambda (Python API backend), and API Gateway. Infrastructure defined in CloudFormation; GitHub Actions syncs updates to S3 and invalidates the CloudFront cache on every push to main. This site runs on it.

WordPress High-Availability Deployment

A highly available WordPress deployment across two Availability Zones: public and private subnets in a custom VPC, an Application Load Balancer fronting EC2 instances in an Auto Scaling group, RDS Multi-AZ for the database, EFS for shared media, NAT Gateways for outbound access from private subnets, Route 53 for DNS, and ACM for TLS.